TEST BANK FOR Data Mining Concepts and Techniques 2nd Edition By Jiawei Han, Micheline Kamber (Solution Manual)

Exam (elaborations) TEST BANK FOR Data Mining Concepts and Techniques 2nd Edition By Jiawei Han, Micheline Kamber (Solution Manual) Data Mining: Concepts and Techniques 2nd Edition Solution Manual Jiawei Han and Micheline Kamber The University of Illinois at Urbana-Champaign °c Morgan Kaufmann, 2006 Note: For Instructors' reference only. Do not copy! Do not distribute! Contents 1 Introduction 3 1.11 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 2 Data Preprocessing 13 2.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 3 Data Warehouse and OLAP Technology: An Overview 31 3.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 4 Data Cube Computation and Data Generalization 41 4.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 5 Mining Frequent Patterns, Associations, and Correlations 53 5.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 6 Classi¯cation and Prediction 69 6.17 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69 7 Cluster Analysis 79 7.13 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 8 Mining Stream, Time-Series, and Sequence Data 91 8.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 9 Graph Mining, Social Network Analysis, and Multirelational Data Mining 103 9.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 10 Mining Object, Spatial, Multimedia, Text, and Web Data 111 10.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 11 Applications and Trends in Data Mining 123 11.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 1 Chapter 1 Introduction 1.11 Exercises 1.1. What is data mining? In your answer, address the following: (a) Is it another hype? (b) Is it a simple transformation of technology developed from databases, statistics, and machine learning? (c) Explain how the evolution of database technology led to data mining. (d) Describe the steps involved in data mining when viewed as a process of knowledge discovery. Answer: Data mining refers to the process or method that extracts or mines" interesting knowledge or patterns from large amounts of data. (a) Is it another hype? Data mining is not another hype. Instead, the need for data mining has arisen due to the wide availability of huge amounts of data and the imminent need for turning such data into useful information and knowledge. Thus, data mining can be viewed as the result of the natural evolution of information technology. (b) Is it a simple transformation of technology developed from databases, statistics, and machine learning? No. Data mining is more than a simple transformation of technology developed from databases, sta- tistics, and machine learning. Instead, data mining involves an integration, rather than a simple transformation, of techniques from multiple disciplines such as database technology, statistics, ma- chine learning, high-performance computing, pattern recognition, neural networks, data visualization, information retrieval, image and signal processing, and spatial data analysis. (c) Explain how the evolution of database technology led to data mining. Database technology began with the development of data collection and database creation mechanisms that led to the development of e®ective mechanisms for data management including data storage and retrieval, and query and transaction processing. The large number of database systems o®ering query and transaction processing eventually and naturally led to the need for data analysis and understanding. Hence, data mining began its development out of this necessity. (d) Describe the steps involved in data mining when viewed as a process of knowledge discovery. The steps involved in data mining when viewed as a process of knowledge discovery are as follows: ² Data cleaning, a process that removes or transforms noise and inconsistent data ² Data integration, where multiple data sources may be combined 3 4 CHAPTER 1. INTRODUCTION ² Data selection, where data relevant to the analysis task are retrieved from the database ² Data transformation, where data are transformed or consolidated into forms appropriate for mining ² Data mining, an essential process where intelligent and e±cient methods are applied in order to extract patterns ² Pattern evaluation, a process that identi¯es the truly interesting patterns representing knowl- edge based on some interestingness measures ² Knowledge presentation, where visualization and knowledge representation techniques are used to present the mined knowledge to the user 1.2. Present an example where data mining is crucial to the success of a business. What data mining functions does this business need? Can they be performed alternatively by data query processing or simple statistical analysis? Answer: A department store, for example, can use data mining to assist with its target marketing mail campaign. Using data mining functions such as association, the store can use the mined strong association rules to determine which products bought by one group of customers are likely to lead to the buying of certain other products. With this information, the store can then mail marketing materials only to those kinds of customers who exhibit a high likelihood of purchasing additional products. Data query processing is used for data or information retrieval and does not have the means for ¯nding association rules. Similarly, simple statistical analysis cannot handle large amounts of data such as those of customer records in a department store. 1.3. Suppose your task as a software engineer at Big-University is to design a data mining system to examine their university course database, which contains the following information: the name, address, and status (e.g., undergraduate or graduate) of each student, the courses taken, and their cumulative grade point average (GPA). Describe the architecture you would choose. What is the purpose of each component of this architecture? Answer: A data mining architecture that can be used for this application would consist of the following major components: ² A database, data warehouse, or other information repository, which consists of the set of databases, data warehouses, spreadsheets, or other kinds of information repositories containing the student and course information. ² A database or data warehouse server, which fetches the relevant data based on the users' data mining requests. ² A knowledge base that contains the domain knowledge used to guide the search or to evaluate the interestingness of resulting patterns. For example, the knowledge base may contain concept hierarchies and metadata (e.g., describing data from multiple heterogeneous sources). ² A data mining engine, which consists of a set of functional modules for tasks such as classi¯cation, association, classi¯cation, cluster analysis, and evolution and deviation analysis. ² A pattern evaluation module that works in tandem with the data mining modules by employing interestingness measures to help focus the search towards interesting patterns. ² A graphical user interface that provides the user with an interactive approach to the data mining system. 1.11. EXERCISES 5 1.4. How is a data warehouse di®erent from a database? How are they similar? Answer: ² Di®erences between a data warehouse and a database: A data warehouse is a repository of informa- tion collected from multiple sources, over a history of time, stored under a uni¯ed schema, and used for data analysis and decision support; whereas a database, is a collection of interrelated data that rep- resents the current status of the stored data. There could be multiple heterogeneous databases where the schema of one database may not agree with the schema of another. A database system supports ad-hoc query and on-line transaction processing. Additional di®erences are detailed in Section 3.1.1 Di®erences between Operational Databases Systems and Data Warehouses. ² Similarities between a data warehouse and a database: Both are repositories of information, storing huge amounts of persistent data. 1.5. Brie°y describe the following advanced database systems and applications: object-relational databases, spatial databases, text databases, multimedia databases, the World Wide Web. Answer: ² An objected-oriented database is designed based on the object-oriented programming paradigm where data are a large number of objects organized into classes and class hierarchies. Each entity in the database is considered as an object. The object contains a set of variables that describe the object, a set of messages that the object can use to communicate with other objects or with the rest of the database system, and a set of methods where each method holds the code to implement a message. ² A spatial database contains spatial-related data, which may be represented in the form of raster or vector data. Raster data consists of n-dimensional bit maps or pixel maps, and vector data are represented by lines, points, polygons or other kinds of processed primitives, Some examples of spatial databases include geographical (map) databases, VLSI chip designs, and medical and satellite images databases. ² A text database is a database that contains text documents or other word descriptions in the form of long sentences or paragraphs, such as product speci¯cations, error or bug reports, warning messages, summary reports, notes, or other documents. ² A multimedia database stores images, audio, and video data, and is used in applications such as picture content-based retrieval, voice-mail systems, video-on-demand systems, the World Wide Web, and speech-based user interfaces. ² The World Wide Web provides rich, world-wide, on-line information services, where data objects are linked together to facilitate interactive access. Some examples of distributed information services associated with the World Wide Web include America Online, Yahoo!, AltaVista, and Prodigy. 1.6. De¯ne each of the following data mining functionalities: characterization, discrimination, association and correlation analysis, classi¯cation, prediction, clustering, and evolution analysis. Give examples of each data mining functionality, using a real-life database that you are familiar with. Answer: ² Characterization is a summarization of the general characteristics or features of a target class of data. For example, the characteristics of students can be produced, generating a pro¯le of all the University ¯rst year computing science students, which may include such information as a high GPA and large number of courses taken. ² Discrimination is a comparison of the general features of target class data objects with the general features of objects from one or a set of contrasting classes. For example, the general features of students with high GPA's may be compared with the general features of students with low GPA's. The resulting 6 CHAPTER 1. INTRODUCTION description could be a general comparative pro¯le of the students such as 75% of the students with high GPA's are fourth-year computing science students while 65% of the students with low GPA's are not. ² Association is the discovery of association rules showing attribute-value conditions that occur fre- quently together in a given set of data. For example, a data mining system may ¯nd association rules like major(X; computing science"") ) owns(X; personal computer") [support = 12%; confidence = 98%] where X is a variable representing a student. The rule indicates that of the students under study, 12% (support) major in computing science and own a personal computer. There is a 98% probability (con¯dence, or certainty) that a student in this group owns a personal computer. ² Classi¯cation di®ers from prediction in that the former constructs a set of models (or functions) that describe and distinguish data classes or concepts, whereas the latter builds a model to predict some missing or unavailable, and often numerical, data values. Their similarity is that they are both tools for prediction: Classi¯cation is used for predicting the class label of data objects and prediction is typically used for predicting missing numerical data values. ² Clustering analyzes data objects without consulting a known class label. The objects are clustered or grouped based on the principle of maximizing the intraclass similarity and minimizing the interclass similarity. Each cluster that is formed can be viewed as a class of objects. Clustering can also facilitate taxonomy formation, that is, the organization of observations into a hierarchy of classes that group similar events together. ² Data evolution analysis describes and models regularities or trends for objects whose behavior changes over time. Although this may include characterization, discrimination, association, classi¯ca- tion, or clustering of time-related data, distinct features of such an analysis include time-series data analysis, sequence or periodicity pattern matching, and similarity-based data analysis. 1.7. What is the di®erence between discrimination and classi¯cation? Between characterization and clustering? Between classi¯cation and prediction? For each of these pairs of tasks, how are they similar? Answer: ² Discrimination di®ers from classi¯cation in that the former refers to a comparison of the general features of target class data objects with the general features of objects from one or a set of contrasting classes, while the latter is the process of ¯nding a set of models (or functions) that describe and distinguish data classes or concepts for the purpose of being able to use the model to predict the class of objects whose class label is unknown. Discrimination and classi¯cation are similar in that they both deal with the analysis of class data objects. ² Characterization di®ers from clustering in that the former refers to a summarization of the general characteristics or features of a target class of data while the latter deals with the analysis of data objects without consulting a known class label. This pair of tasks is similar in that they both deal with grouping together objects or data that are related or have high similarity in comparison to one another. ² Classi¯cation di®ers from prediction in that the former is the process of ¯nding a set of models (or functions) that describe and distinguish data class or concepts while the latter predicts missing or unavailable, and often numerical, data values. This pair of tasks is similar in that they both are tools for prediction: Classi¯cation is used for predicting the class label of data objects and prediction is typically used for predicting missing numerical data values. 1.11. EXERCISES 7 1.8. Based on your observation, describe another possible kind of knowledge that needs to be discovered by data mining methods but has not been listed in this chapter. Does it require a mining methodology that is quite di®erent from those outlined in this chapter? Answer: There is no standard answer for this question and one can judge the quality of an answer based on the freshness and quality of the proposal. For example, one may propose partial periodicity as a new kind of knowledge, where a pattern is partial periodic if only some o®sets of a certain time period in a time series demonstrate some repeating behavior. 1.9. List and describe the ¯ve primitives for specifying a data mining task. Answer: The ¯ve primitives for specifying a data-mining task are: ² Task-relevant data: This primitive speci¯es the data upon which mining is to be performed. It involves specifying the database and tables or data warehouse containing the relevant data, conditions for selecting the relevant data, the relevant attributes or dimensions for exploration, and instructions regarding the ordering or grouping of the data retrieved. ² Knowledge type to be mined: This primitive speci¯es the speci¯c data mining function to be performed, such as characterization, discrimination, association, classi¯cation, clustering, or evolution analysis. As well, the user can be more speci¯c and provide pattern templates that all discovered patterns must match. These templates, or metapatterns (also called metarules or metaqueries), can be used to guide the discovery process. ² Background knowledge: This primitive allows users to specify knowledge they have about the domain to be mined. Such knowledge can be used to guide the knowledge discovery process and evaluate the patterns that are found. Concept hierarchies and user beliefs regarding relationships in the data are forms of background knowledge. ² Pattern interestingness measure: This primitive allows users to specify functions that are used to separate uninteresting patterns from knowledge and may be used to guide the mining process, as well as to evaluate the discovered patterns. This allows the user to con¯ne the number of uninteresting patterns returned by the process, as a data mining process may generate a large number of patterns. Interestingness measures can be speci¯ed for such pattern characteristics as simplicity, certainty, utility and novelty. ² Visualization of discovered patterns: This primitive refers to the form in which discovered patterns are to be displayed. In order for data mining to be e®ective in conveying knowledge to users, data mining systems should be able to display the discovered patterns in multiple forms such as rules, tables, cross tabs (cross-tabulations), pie or bar charts, decision trees, cubes or other visual representations. 1.10. Describe why concept hierarchies are useful in data mining. Answer: Concept hierarchies de¯ne a sequence of mappings from a set of lower-level concepts to higher-level, more general concepts and can be represented as a set of nodes organized in a tree, in the form of a lattice, or as a partial order. They are useful in data mining because they allow the discovery of knowledge at multiple levels of abstraction and provide the structure on which data can be generalized (rolled-up) or specialized (drilled-down). Together, these operations allow users to view the data from di®erent perspectives, gaining further insight into relationships hidden in the data. Generalizing has the advantage of compressing the data set, and mining on a compressed data set will require fewer I/O operations. This will be more e±cient than mining on a large, uncompressed data set. 8 CHAPTER 1. INTRODUCTION 1.11. Outliers are often discarded as noise. However, one person's garbage could be another's treasure. For example, exceptions in credit card transactions can help us detect the fraudulent use of credit cards. Taking fraudulence detection as an example, propose two methods that can be used to detect outliers and discuss which one is more reliable. Answer: ² Using clustering techniques: After clustering, the di®erent clusters represent the di®erent kinds of data (transactions). The outliers are those data points that do not fall into any cluster. Among the various kinds of clustering methods, density-based clustering may be the most e®ective. Clustering is detailed in Chapter 8. ² Using prediction (or regression) techniques: Constructed a probability (regression) model based on all of the data. If the predicted value for a data point di®ers greatly from the given value, then the given value may be consider an outlier. Outlier detection based on clustering techniques may be more reliable. Because clustering is unsupervised, we do not need to make any assumptions regarding the data distribution (e.g., density-based methods). In contrast, regression (prediction) methods require us to make some assumptions of the data distribution, which may be inaccurate due to insu±cient data. 1.12. Recent applications pay special attention to spatiotemporal data streams. A spatiotemporal data stream contains spatial information that changes over time, and is in the form of stream data, i.e., the data °ow in-and-out like possibly in¯nite streams. (a) Present three application examples of spatiotemporal data streams. (b) Discuss what kind of interesting knowledge can be mined from such data streams, with limited time and resources. (c) Identify and discuss the major challenges in spatiotemporal data mining. (d) Using one application example, sketch a method to mine one kind of knowledge from such stream data e±ciently. Answer: (a) Present three application examples of spatiotemporal data streams. i. Sequences of sensor images of a geographical region along time. ii. The climate images from satellites. iii. Data that describe the evolution of natural phenomena, such as forest coverage, forest ¯re, and so on. (b) Discuss what kind of interesting knowledge can be mined from such data streams, with limited time and resources. The knowledge that can be mined from spatiotemporal data streams really depends on the application. However, one unique type of knowledge about stream data is the patterns of spatial change with respect to the time. For example, the changing of the tra±c status of several highway junctions in a city, from the early morning to rush hours and back to o®-peak hours, can show clearly where the tra±c comes from and goes to and hence, would help the tra±c o±cer plan e®ective alternative lanes in order to reduce the tra±c load. As another example, a sudden appearance of a point in the spectrum space image may indicate that a new planet is being formed. The changing of humidity, temperature, and pressure in climate data may reveal patterns of how a new typhoon is created. (c) Identify and discuss the major challenges in spatiotemporal data mining. One major challenge is how to deal with the continuing large-scale data. Since the data keep °owing in and each snapshot of data is usually huge (e.g., the spectrum image of space), it is [old: almost][new: 1.11. EXERCISES 9 often] impossible to store all of the data. Some aggregation or compression techniques may have to be applied, and old raw data may have to be dropped. Mining under such aggregated (or lossy) data is challenging. In addition, some patterns may occur with respect to a long time period, but it may not be possible to keep the data for such a long duration. Thus, these patterns may not be uncovered. The spatial data sensed may not be so accurate, so the algorithms must have high tolerance with respect to noise. (d) Using one application example, sketch a method to mine one kind of knowledge from such stream data e±ciently. Take mining space images as the application. We seek to observe whether any new planet is being created or any old planet is disappearing. This is a change detection problem. Since the image frames keep coming, that is, f1; : : : ; ft; ft+1; : : :, we can simplify the overall problem to that of detecting whether any planet appears or disappears between two consecutive image frames, ft and ft+1. The algorithm can be sketched as follows. For each incoming frame, ft+1, compare it with the previous frame, ft. i. Match the planets in ft+1 with ft. ii. Detect whether there are any unmatched" planets (where a planet in one of the two frames does not occur in the other). iii. If yes, report a planet appearance (if an unmatched planet appears in the new frame) or a planet disappearance (if an unmatched planet appears in the old frame). In fact, matching between two frames may not be easy because the earth is rotating and thus, the sensed data may have slight variations. Some advanced techniques from image processing may be applied. The overall skeleton of the algorithm is simple. Each new incoming image frame is only compared with the previous one, satisfying the time and resource constraint. The reported change would be useful since it is [old: almost impossible][new: infeasible] for astronomers to dig into every frame to detect whether a new planet has appeared or an old one has disappeared. 1.13. Describe the di®erences between the following approaches for the integration of a data mining system with a database or data warehouse system: no coupling, loose coupling, semitight coupling, and tight coupling. State which approach you think is the most popular, and why. Answer: The di®erences between the following architectures for the integration of a data mining system with a database or data warehouse system are as follows. ² No coupling: The data mining system uses sources such as °at ¯les to obtain the initial data set to be mined since no database system or data warehouse system functions are implemented as part of the process. Thus, this architecture represents a poor design choice. ² Loose coupling: The data mining system is not integrated with the database or data warehouse system beyond their use as the source of the initial data set to be mined and possible use in storage of the results. Thus, this architecture can take advantage of the °exibility, e±ciency, and features (such as indexing) that the database and data warehousing systems may provide. However, it is di±cult for loose coupling to achieve high scalability and good performance with large data sets as many such systems are memory-based. ² Semitight coupling: Some of the data mining primitives, such as aggregation, sorting, or precompu- tation of statistical functions, are e±ciently implemented in the database or data warehouse system for use by the data mining system during mining-query processing. Also, some frequently used intermedi- ate mining results can be precomputed and stored in the database or data warehouse system, thereby enhancing the performance of the data mining system. 10 CHAPTER 1. INTRODUCTION ² Tight coupling: The database or data warehouse system is fully integrated as part of the data mining system and thereby provides optimized data mining query processing. Thus, the data mining subsystem is treated as one functional component of an information system. This is a highly desirable architecture as it facilitates e±cient implementations of data mining functions, high system performance, and an integrated information processing environment. From the descriptions of the architectures provided above, it can be seen that tight coupling is the best alternative without respect to technical or implementation issues. However, as much of the technical in- frastructure needed in a tightly coupled system is still evolving, implementation of such a system is non- trivial. Therefore, the most popular architecture is currently semitight coupling as it provides a compromise between loose and tight coupling. 1.14. Describe three challenges to data mining regarding data mining methodology and user interaction issues. Answer: Challenges to data mining regarding data mining methodology and user interaction issues include the following: mining di®erent kinds of knowledge in databases, interactive mining of knowledge at multiple levels of abstraction, incorporation of background knowledge, data mining query languages and ad hoc data mining, presentation and visualization of data mining results, handling noisy or incomplete data, and pattern evaluation. Below are the descriptions of the ¯rst three challenges mentioned: ² Mining di®erent kinds of knowledge in databases: Di®erent users are interested in di®erent kinds of knowledge and will require a wide range of data analysis and knowledge discovery tasks such as data characterization, discrimination, association, classi¯cation, clustering, trend and deviation analysis, and similarity analysis. Each of these tasks will use the same database in di®erent ways and will require di®erent data mining techniques. ² Interactive mining of knowledge at multiple levels of abstraction: Interactive mining, with the use of OLAP operations on a data cube, allows users to focus the search for patterns, providing and re¯ning data mining requests based on returned results. The user can then interactively view the data and discover patterns at multiple granularities and from di®erent angles. ² Incorporation of background knowledge: Background knowledge, or information regarding the domain under study such as integrity constraints and deduction rules, may be used to guide the discovery process and allow discovered patterns to be expressed in concise terms and at di®erent levels of abstraction. This helps to focus and speed up a data mining process or judge the interestingness of discovered patterns. 1.15. What are the major challenges of mining a huge amount of data (such as billions of tuples) in comparison with mining a small amount of data (such as a few hundred tuple data set)? Answer: One challenge to data mining regarding performance issues is the e±ciency and scalability of data mining algorithms. Data mining algorithms must be e±cient and scalable in order to e®ectively extract information from large amounts of data in databases within predictable and acceptable running times. Another challenge is the parallel, distributed, and incremental processing of data mining algorithms. The need for parallel and distributed data mining algorithms has been brought about by the huge size of many databases, the wide distribution of data, and the computational complexity of some data mining methods. Due to the high cost of some data mining processes, incremental data mining algorithms incorporate database updates without the need to mine the entire data again from scratch. 1.16. Outline the major research challenges of data mining in one speci¯c application domain, such as stream/sensor data analysis, spatiotemporal data analysis, or bioinformatics. 1.11. EXERCISES 11 Answer: Students must research their answers for this question. Major data mining challenges for two applications, that of data streams and bioinformatics, are addressed here. Data Stream Data stream analysis presents multiple challenges. First, data streams are continuously °owing in and out as well as changing dynamically. The data analysis system that will successfully take care of this type of data needs to be in real time and able to adapt to changing patterns that might emerge. Another major challenge is that the size of stream data can be huge or even in¯nite. Because of this size, only a single or small number of scans are typically allowed. For further details on mining data stream, please consult Chapter 8. Bioinformatics The ¯eld of bioinformatics encompasses many other sub¯elds like genomics, proteomics, molecular biology, and chemi-informatics. Each of these individual sub¯elds has many research challenges. Some of the major challenges of data mining in the ¯eld of bioinformatics are outlined as follows. (Due to limitations of space, some of the terminology used here may not be explained.) ² Data explosion: Biological data are growing at an exponential rate. It has been estimated that genomic and proteomic data are doubling every 12 months. Most of these data are scattered around in unstructured and nonstandard forms in various di®erent databases throughout the research community. Many of the biological experiments do not yield exact results and are prone to errors because it is very di±cult to model exact biological conditions and processes. For example, the structure of a protein is not rigid and is dependent on its environment. Hence, the structures determined by nuclear magnetic resonance (NMR) or crystallography experiments may not represent the exact structure of the protein. Since these experiments are performed in parallel by many institutions and scientists, they may each yield slightly di®erent structures. The consolidation and validation of these con°icting data is a di±cult challenge. Some research labs have come up with public domain repositories of data, such as the Protein Data Bank (PDB). These have become very popular in the past few years. However, due to concerns of Intellectual Property, a great deal of useful biological information is buried in proprietary databases within large pharmaceutical companies. ² Text mining from research publications/repositories: Most of the data generated in the biological research community is from experiments. Most of the results are published, but they are seldom recorded in databases with the experiment details (who, when, how, etc.). Hence, a great deal of useful information is buried in published and unpublished literature. This has given rise to the need for the development of text mining systems. For example, many experimental results regarding protein interactions have been published. Mining this information may provide crucial insight into biological pathways and help predict potential interactions. The extraction and development of domain-speci¯c ontologies is also another related research challenge. ² Mining large databases of compounds/molecules: The major steps in a drug discovery phase include target identi¯cation, target validation, lead discovery, and lead optimization. The most time- consuming step is the lead discovery phase. In this step, large databases of compounds are needed to be mined to identify potential lead candidates that will suitably interact with the potential target. Currently, due to the lack of e®ective data mining systems, this step involves many trial-and-error iterations of wet lab or protein assay experiments. These experiments are highly time-consuming and costly. Hence, one of the current challenges in bioinformatics includes the development of intelligent and computational data mining systems that can eliminate false positives and generate more true positives before the wet lab experimentation stage. This task is particularly challenging, because it involves the development of a mining/screening system that can identify compounds that can dock1 well with the target compound. The docking problem is an especially tricky problem, because it is governed by many physical interactions at the molecular level. Some progress has been made regarding pair- wise docking, where large time-consuming Molecular Dynamics (MD) simulation-based optimization 1Docking is a term used for computational schemes that attempt to ¯nd the best matching" between two molecules: a receptor and a ligand. 12 CHAPTER 1. INTRODUCTION methods can predict docking to a good degree of success. The main problem is the large solution space generated by the complex interactions at the molecular level. The molecular docking problem remains a fairly unsolved problem. The major research challenges in the mining of these interactions include the development of fast and accurate algorithms for screening and ranking these compounds/molecules based on their ability to interact with a given compound/molecule. Other related research areas include protein classi¯cation systems based on structure and function. ² Pattern analysis and classi¯cation of microarray data: A great deal of progress has been made in the past decade regarding the development of algorithms for the analysis of genomic data. Statistical and other methods are available. A large research community in data mining is focusing on adopting these pattern analysis and classi¯cation methods for mining microarray and gene expression data. Chapter 2 Data Preprocessing 2.8 Exercises 2.1. Data quality can be assessed in terms of accuracy, completeness, and consistency. Propose two other dimensions of data quality. Answer: Other dimensions that can be used to assess the quality of data include timeliness, believability, value added, interpretability and accessability, described as follows: ² Timeliness: Data must be available within a time frame that allows it to be useful for decision making. ² Believability: Data values must be within the range of possible results in order to be useful for decision making. ² Value added: Data must provide additional value in terms of information that o®sets the cost of collecting and accessing it. ² Interpretability: Data must not be so complex that the e®ort to understand the information it provides exceeds the bene¯t of its analysis. ² Accessability: Data must be accessible so that the e®ort to collect it does not exceed the bene¯t from its use. 2.2. Suppose that the values for a given set of data are grouped into intervals. The intervals and corresponding frequencies are as follows. age frequency 1-5 200 5-15 450 15-20 300 50-80 700 80-110 44 Compute an approximate median value for the data. Answer: Using Equation (2.3), we have L1 = 20, N= 3194, ( P freq)l = 950, freqmedian = 1500, width = 30, median = 32.94 years. 13 14 CHAPTER 2. DATA PREPROCESSING 2.3. Give three additional commonly used statistical measures (i.e., not illustrated in this chapter) for the characterization of data dispersion, and discuss how they can be computed e±ciently in large databases. Answer: Data dispersion, also known as variance analysis, is the degree to which numeric data tend to spread and can be characterized by such statistical measures as mean deviation, measures of skewness, and the coe±cient of variation. The mean deviation is de¯ned as the arithmetic mean of the absolute deviations from the means and is calculated as: mean deviation = PN i=1 jx ¡ ¹xj N ; (2.1) where ¹x is the arithmetic mean of the values and N is the total number of values. This value will be greater for distributions with a larger spread. A common measure of skewness is: ¹x ¡ mode s ; (2.2) which indicates how far (in standard deviations, s) the mean (¹x) is from the mode and whether it is greater or less than the mode. The coe±cient of variation is the standard deviation expressed as a percentage of the arithmetic mean and is calculated as: coe±cient of variation = s ¹x £ 100 (2.3) The variability in groups of observations with widely di®ering means can be compared using this measure. Note that all of the input values used to calculate these three statistical measures are algebraic measures. Thus, the value for the entire database can be e±ciently calculated by partitioning the database, computing the values for each of the separate partitions, and then merging theses values into an algebraic equation that can be used to calculate the value for the entire database. The measures of dispersion described here were obtained from: Statistical Methods in Research and Produc- tion, fourth ed., edited by Owen L. Davies and Peter L. Goldsmith, Hafner Publishing Company, NY:NY, 1972. 2.4. Suppose that the data for analysis includes the attribute age. The age values for the data tuples are (in increasing order) 13, 15, 16, 16, 19, 20, 20, 21, 22, 22, 25, 25, 25, 25, 30, 33, 33, 35, 35, 35, 35, 36, 40, 45, 46, 52, 70. (a) What is the mean of the data? What is the median? (b) What is the mode of the data? Comment on the data's modality (i.e., bimodal, trimodal, etc.). (c) What is the midrange of the data? (d) Can you ¯nd (roughly) the ¯rst quartile (Q1) and the third quartile (Q3) of the data? (e) Give the ¯ve-number summary of the data. (f) Show a boxplot of the data. (g) How is a quantile-quantile plot di®erent from a quantile plot? 2.8. EXERCISES 15 Answer: (a) What is the mean of the data? What is the median? The (arithmetic) mean of the data is: ¹x = 1 N XN i=1 xi = 809=27 = 30 (Equation 2.1). The median (middle value of the ordered set, as the number of values in the set is odd) of the data is: 25. (b) What is the mode of the data? Comment on the data's modality (i.e., bimodal, trimodal, etc.). This data set has two values that occur with the same highest frequency and is, therefore, bimodal. The modes (values occurring with the greatest frequency) of the data are 25 and 35. (c) What is the midrange of the data? The midrange (average of the largest and smallest values in the data set) of the data is: (70+13)=2 = 41:5 (d) Can you ¯nd (roughly) the ¯rst quartile (Q1) and the third quartile (Q3) of the data? The ¯rst quartile (corresponding to the 25th percentile) of the data is: 20. The third quartile (corre- sponding to the 75th percentile) of the data is: 35. (e) Give the ¯ve-number summary of the data. The ¯ve number summary of a distribution consists of the minimum value, ¯rst quartile, median value, third quartile, and maximum value. It provides a good summary of the shape of the distribution and for this data is: 13, 20, 25, 35, 70. (f) Show a boxplot of the data. (Omitted here. Please refer to Figure 2.3 of the textbook.) (g) How is a quantile-quantile plot di®erent from a quantile plot? A quantile plot is a graphical method used to show the approximate percentage of values below or equal to the independent variable in a univariate distribution. Thus, it displays quantile information for all the data, where the values measured for the independent variable are plotted against their corresponding quantile. A quantile-quantile plot however, graphs the quantiles of one univariate distribution against the corre- sponding quantiles of another univariate distribution. Both axes display the range of values measured for their corresponding distribution, and points are plotted that correspond to the quantile values of the two distributions. A line (y = x) can be added to the graph along with points representing where the ¯rst, second and third quantiles lie to increase the graph's informational value. Points that lie above such a line indicate a correspondingly higher value for the distribution plotted on the y-axis than for the distribution plotted on the x-axis at the same quantile. The opposite e®ect is true for points lying below this line. 2.5. In many applications, new data sets are incrementally added to the existing large data sets. Thus an important consideration for computing descriptive data summary is whether a measure can be computed e±ciently in incremental manner. Use count, standard deviation, and median as examples to show that a distributive or algebraic measure facilitates e±cient incremental computation, whereas a holistic measure does not. Answer: ² Count: The current count can be stored as a value, and when x number of new values are added, we can easily update count with count + x. This is a distributive measure and is easily updated for incremental additions. ² Standard deviation: If we store the sum of the squared existing values and the count of the existing values, we can easily generate the new standard deviation using the formula provided in the book. We simply need to calculate the squared sum of the new numbers, add that to the existing squared sum, update the count of the numbers, and plug that into the calculation to obtain the new standard deviation. All of this is done without looking at the whole data set and is thus easy to compute. 16 CHAPTER 2. DATA PREPROCESSING ² Median: To accurately calculate the median, we have to look at every value in the dataset. When we add a new value or values, we have to sort the new set and then ¯nd the median based on that new sorted set. This is much harder and thus makes the incremental addition of new values di±cult. 2.6. In real-world data, tuples with missing values for some attributes are a common occurrence. Describe various methods for handling this problem. Answer: The various methods for handling the problem of missing values in data tuples include: (a) Ignoring the tuple: This is usually done when the class label is missing (assuming the mining task involves classi¯cation or description). This method is not very e®ective unless the tuple contains several attributes with missing values. It is especially poor when the percentage of missing values per attribute varies considerably. (b) Manually ¯lling in the missing value: In general, this approach is time-consuming and may not be a reasonable task for large data sets with many missing values, especially when the value to be ¯lled in is not easily determined. (c) Using a global constant to ¯ll in the missing value: Replace all missing attribute values by the same constant, such as a label like Unknown," or ¡1. If missing values are replaced by, say, Unknown," then the mining program may mistakenly think that they form an interesting concept, since they all have a value in common | that of Unknown." Hence, although this method is simple, it is not recommended. (d) Using the attribute mean for quantitative (numeric) values or attribute mode for categor- ical (nominal) values: For example, suppose that the average income of AllElectronics customers is $28,000. Use this value to replace any missing values for income. (e) Using the attribute mean for quantitative (numeric) values or attribute mode for categor- ical (nominal) values, for all samples belonging to the same class as the given tuple: For example, if classifying customers according to credit risk, replace the missing value with the average income value for customers in the same credit risk category as that of the given tuple. (f) Using the most probable value to ¯ll in the missing value: This may be determined with regression, inference-based tools using Bayesian formalism, or decision tree induction. For example, using the other customer attributes in the data set, we can construct a decision tree to predict the missing values for income. 2.7. Using the data for age given in Exercise 2.4, answer the following. (a) Use smoothing by bin means to smooth the above data, using a bin depth of 3. Illustrate your steps. Comment on the e®ect of this technique for the given data. (b) How might you determine outliers in the data? (c) What other methods are there for data smoothing? Answer: (a) Use smoothing by bin means to smooth the above data, using a bin depth of 3. Illustrate your steps. Comment on the e®ect of this technique for the given data. The following steps are required to smooth the above data using smoothing by bin means with a bin depth of 3. ² Step 1: Sort the data. (This step is not required here as the data are already sorted.) ² Step 2: Partition the data into equal-frequency bins of size 3. 2.8. EXERCISES 17 Bin 1: 13, 15, 16 Bin 2: 16, 19, 20 Bin 3: 20, 21, 22 Bin 4: 22, 25, 25 Bin 5: 25, 25, 30 Bin 6: 33, 33, 35 Bin 7: 35, 35, 35 Bin 8: 36, 40, 45 Bin 9: 46, 52, 70 ² Step 3: Calculate the arithmetic mean of each bin. ² Step 4: Replace each of the values in each bin by the arithmetic mean calculated for the bin. Bin 1: 142/3, 142/3, 142/3 Bin 2: 181/3, 181/3, 181/3 Bin 3: 21, 21, 21 Bin 4: 24, 24, 24 Bin 5: 262/3, 262/3, 262/3 Bin 6: 332/3, 332/3, 332/3 Bin 7: 35, 35, 35 Bin 8: 401/3, 401/3, 401/3 Bin 9: 56, 56, 56 (b) How might you determine outliers in the data? Outliers in the data may be detected by clustering, where similar values are organized into groups, or clusters". Values that fall outside of the set of clusters may be considered outliers. Alternatively, a combination of computer and human inspection can be used where a predetermined data distribution is implemented to allow the computer to identify possible outliers. These possible outliers can then be veri¯ed by human inspection with much less e®ort than would be required to verify the entire initial data set. (c) What other methods are there for data smoothing? Other methods that can be used for data smoothing include alternate forms of binning such as smooth- ing by bin medians or smoothing by bin boundaries. Alternatively, equal-width bins can be used to implement any of the forms of binning, where the interval range of values in each bin is constant. Methods other than binning include using regression techniques to smooth the data by ¯tting it to a function such as through linear or multiple regression. Classi¯cation techniques can be used to imple- ment concept hierarchies that can smooth the data by rolling-up lower level concepts to higher-level concepts. 2.8. Discuss issues to consider during data integration. Answer: Data integration involves combining data from multiple sources into a coherent data store. Issues that must be considered during such integration include: ² Schema integration: The metadata from the di®erent data sources must be integrated in order to match up equivalent real-world entities. This is referred to as the entity identi¯cation problem. ² Handling redundant data: Derived attributes may be redundant, and inconsistent attribute naming may also lead to redundancies in the resulting data set. Duplications at the tuple level may occur and thus need to be detected and resolved. ² Detection and resolution of data value con°icts: Di®erences in representation, scaling, or encod- ing may cause the same real-world entity attribute values to di®er in the data sources being integrated. 2.9. Suppose a hospital tested the age and body fat data for 18 randomly selected adults with the following result age %fat 9.5 26.5 7.8 17.8 31.4 25.9 27.4 27.2 31.2 age %fat 34.6 42.5 28.8 33.4 30.2 34.1 32.9 41.2 35.7 (a) Calculate the mean, median and standard deviation of age and %fat. (b) Draw the boxplots for age and %fat. (c) Draw a scatter plot and a q-q plot based on these two variables. 18 CHAPTER 2. DATA PREPROCESSING (d) Normalize the two variables based on z-score normalization. (e) Calculate the correlation coe±cient (Person's product moment coe±cient). Are these two variables positively or negatively correlated? Answer: (a) Calculate the mean, median and standard deviation of age and %fat. For the variable age the mean is 46:44, the median is 51, and the standard deviation is 12:85. For the variable %fat the mean is 28:78, the median is 30:7, and the standard deviation is 8:99. (b) Draw the boxplots for age and %fat. See Figure 2.1. 1 25 30 35 40 45 50 55 60 Values age 1 10 15 20 25 30 35 40 Values %fat Figure 2.1: A boxplot of the variables age and %fat in Exercise 2.9. (c) Draw a scatter plot and a q-q plot based on these two variables. See Figure 2.2. 5 10 15 20 25 30 35 40 45 age fat qq plot 5 10 15 20 25 30 35 40 45 scatter plot age fat Figure 2.2: A q-q plot and a scatter plot of the variables age and %fat in Exercise 2.9. (d) Normalize the two variables based on z-score normalization. 2.8. EXERCISES 19 age z-age -1.83 -1.83 -1.51 -1.51 -0.58 -0.42 0.04 0.20 0.28 %fat 9.5 26.5 7.8 17.8 31.4 25.9 27.4 27.2 31.2 z-%fat -2.14 -0.25 -2.33 -1.22 0.29 -0.32 -0.15 -0.18 0.27 age z-age 0.43 0.59 0.59 0.74 0.82 0.90 0.90 1.06 1.13 %fat 34.6 42.5 28.8 33.4 30.2 34.1 32.9 41.2 35.7 z-%fat 0.65 1.53 0.0 0.51 0.16 0.59 0.46 1.38 0.77 (e) Calculate the correlation coe±cient (Pearson's product moment coe±cient). Are these two variables positively or negatively correlated? The correlation coe±cient is 0:82. The variables are positively correlated. 2.10. What are the value ranges of the following normalization methods? (a) min-max normalization (b) z-score normalization (c) normalization by decimal scaling Answer: (a) min-max normalization The rage is [new min, new max] (b) z-score normalization The range is [(old min¡mean)=stddev; (old max¡mean)=stddev]. In general the range for all possible data sets is (¡1;+1). (c) normalization by decimal scaling The range is (¡1:0; 1:0). 2.11. Use the two methods below to normalize the following group of data: 200; 300; 400; 600; 1000 (a) min-max normalization by setting min = 0 and max = 1 (b) z-score normalization Answer: (a) min-max normalization by setting min = 0 and max = 1 original data [0,1] normalized 0 0.125 0.25 0.5 1 (b) z-score normalization original data z-score -1.06 -0.7 -0.35 0.35 1.78 2.12. Using the data for age given in Exercise 2.4, answer the following: (a) Use min-max normalization to transform the value 35 for age onto the range [0:0; 1:0]. 20 CHAPTER 2. DATA PREPROCESSING (b) Use z-score normalization to transform the value 35 for age, where the standard deviation of age is 12.94 years. (c) Use normalization by decimal scaling to transform the value 35 for age. (d) Comment on which method you would prefer to use for the given data, giving reasons as to why. Answer: (a) Use min-max normalization to transform the value 35 for age onto the range [0:0; 1:0]. For readability, let A be the attribute age. Using Equation (2.11) with minA = 13, maxA = 70, new minA = 0; new maxA = 1:0, then v = 35 is transformed to v0 = 0:39. (b) Use z-score normalization to transform the value 35 for age, where the standard deviation of age is 12.94 years. Using Equation (2.12) where A = 809/27 = 29.96 and ¾A = 12.94, then v = 35 is transformed to v0 = 0:39. (c) Use normalization by decimal scaling to transform the value 35 for age. Using Equation (2.13) where j = 2, v = 35 is transformed to v0 = 0:35. (d) Comment on which method you would prefer to use for the given data, giving reasons as to why. Given the data, one may prefer decimal scaling for normalization because such a transformation would maintain the data distribution and be intuitive to interpret, while still allowing mining on speci¯c age groups. Min-max normalization has the undesired e®ect of not permitting any future values to fall outside of the current minimum and maximum values without encountering an out of bounds error". As such values may be present in future data, this method is less appropriate. [new Both min-max and Z-score normalization change the original data quite a bit.] Z-score normalization transforms values into measures that represent their distance from the mean, in terms of standard deviations. This type of transformation may not be as intuitive to the user in comparison with decimal scaling. 2.13. Use a °ow chart to summarize the following procedures for attribute subset selection: (a) stepwise forward selection (b) stepwise backward elimination (c) a combination of forward selection and backward elimination Answer: (a) Stepwise forward selection See Figure 2.3. (b) Stepwise backward elimination See Figure 2.4. (c) A combination of forward selection and backward elimination See Figure 2.5. 2.14. Suppose a group of 12 sales price records has been sorted as follows: 5; 10; 11; 13; 15; 35; 50; 55; 72; 92; 204; 215: Partition them into three bins by each of the following methods. (a) equal-frequency partitioning (b) equal-width partitioning 2.8. EXERCISES 21 Figure 2.3: Stepwise forward selection. (c) clustering Answer: (a) equal-frequency partitioning bin 1 5,10,11,13 bin 2 15,35,50,55 bin 3 72,92,204,215 (b) equal-width partitioning The width of each interval is (215 ¡ 5)=3 = 70. bin 1 5,10,11,13,15,35,50,55,72 bin 2 92 bin 3 204,215 (c) clustering We will use a simple clustering technique: partition the data along the 2 biggest gaps in the data. bin 1 5,10,11,13,15 bin 2 35,50,55,72,92 bin 3 204,215 2.15. Using the data for age given in Exercise 2.4, (a) Plot an equal-width histogram of width 10. (b) Sketch examples of each of the following sampling techniques: SRSWOR, SRSWR, cluster sampling, strati¯ed sampling. Use samples of size 5 and the strata youth", middle-aged", and senior". Answer: 22 CHAPTER 2. DATA PREPROCESSING Figure 2.4: Stepwise backward elimination. (a) Plot an equiwidth histogram of width 10. See Figure 2.6. (b) Sketch examples of each of the following sampling techniques: SRSWOR, SRSWR, cluster sampling, strati¯ed sampling. Use samples of size 5 and the strata young", middle-aged", and senior". See Figure 2.7. 2.16. [Contributed by Chen Chen] The median is one of the most important holistic measures in data analysis. Propose several methods for median approximation. Analyze their respective complexity under di®erent parameter settings and decide to what extent the real value can be approximated. Moreover, suggest a heuristic strategy to balance between accuracy and complexity and then apply it to all methods you have given. Answer: This question can be dealt with either theoretically or empirically, but doing some experiments to get the result is perhaps more interesting. Given are some data sets sampled from di®erent distributions, e.g., uniform, Gaussian, exponential, and gamma. (The former two distributions are symmetric, whereas the latter two are skewed). For example, if using Equation (2.3) to do approximation as proposed in the chapter, the most straightforward approach is to partition all of the data into k equal-length intervals. median = L1 + (N=2 ¡ ( P freq)l freqmedian )width; (2.3) where L1 is the lower boundary of the median interval, N is the number of values in the entire data set, ( P freq)l is the sum of the frequencies of all of the intervals that are lower than the median interval, freqmedian is the frequency of the median interval, and width is the width of the median interval. Obviously, the error incurred will be decreased as k becomes larger; however, the time used in the whole procedure will also increase. The product of error made and time used are good optimality measures. From this point, we can perform many tests for each type of distributions (so that the result won't be dominated 2.8. EXERCISES 23 Figure 2.5: A combination of forward selection and backward elimination. by randomness) and ¯nd the k giving the best trade-o®. In practice, this parameter value can be chosen to improve system performance. There are also other approaches for median approximation. The student may suggest a few, analyze the best trade-o® point, and compare the results from the di®erent approaches. A possible such approach is as follows: Hierarchically divide the whole data set into intervals: ¯rst, divide it into k regions and ¯nd the region in which the median resides; second, divide this particular region into k subregions, ¯nd the subregion in which the median resides; . . . . This iterates until the width of the subregion reaches a prede¯ned threshold, and then the median approximation formula as above stated is applied. In this way, we can con¯ne the median to a smaller area without globally partitioning all of data into shorter intervals, which would be expensive. (The cost is proportional to the number of intervals.) 2.17. [Contributed by Deng Cai] It is important to de¯ne or select similarity measures in data analysis. However, there is no commonly accepted subjective similarity measure. Using di®erent similarity measures may deduce di®erent results. Nonetheless, some apparently di®erent similarity measures may be equivalent after some transformation. Suppose we have the following two-dimensional data set: A1 A2 x1 1.5 1.7 x2 2 1.9 x3 1.6 1.8 x4 1.2 1.5 x5 1.5 1.0 (a) Consider the data as two-dimensional data points. Given a new data point, x = (1:4; 1:6) as a query, rank the database points based on similarity with the query using (1) Euclidean distance (Equation 7.5), and (2) cosine similarity (Equation 7.16). (b) Normalize the data set to make the norm of each data point equal to 1. Use Euclidean distance on the transformed data to rank the data points. Answer: 24 CHAPTER 2. DATA PREPROCESSING Figure 2.6: An equiwidth histogram of width 10 for age. (a) Consider the data as two-dimensional data points. Given a new data point, x = (1:4; 1:6) as a query, rank the database points based on similarity with the query using (1) Euclidean distance (Equation 7.5), and (2) cosine similarity (Equation 7.16). The Euclidean distance of two n-dimensional vectors, x and y, is de¯ned as: 2 pPn i=1 (xi ¡ yi)2. The cosine similarity of x and y is de¯ned as: xt¢y jjxjjjjyjj , where xt is a transposition of vector x, jjxjj is the Euclidean norm of vector x,1 and jjyjj is the Euclidean norm of vector y. Using these de¯nitions we obtain the distance from each point to the query point. x1 x2 x3 x4 x5 Euclidean distance 0.14 0.67 0.28 0.22 0.61 Cosine similarity 0.9999 0.9957 0.9999 0.9990 0.9653 Based on the Euclidean distance, the ranked order is x1; x4; x3; x5; x2. Based on the cosine similarity, the order is x1; x3; x4; x2; x5. (b) Normalize the data set to make the norm of each data point equal to 1. Use Euclidean distance on the transformed data to rank the data points. After normalizing the data we have: x x1 x2 x3 x4 x5 0.6585 0.6616 0.7250 0.6644 0.6247 0.8321 0.7526 0.7498 0.6887 0.7474 0.7809 0.5547 The new Euclidean distance is: x1 x2 x3 x4 x5 Euclidean distance 0.0041 0.0922 0.0078 0.0441 0.2632 1The Euclidean normal of vector x = (x1; x2; : : : ; xn) is de¯ned as q x21 + x22 + : : : + x2 n. Conceptually, it is the length of the vector. 2.8. EXERCISES 25 Tuples T1 13 T2 15 T3 16 T4 16 T5 19 T6 20 T7 20 T8 21 T9 22 T10 22 T11 25 T12 25 T13 25 T14 25 T15 30 T16 33 T17 33 T18 33 T19 33 T20 35 T21 35 T22 36 T23 40 T24 45 T25 46 T26 52 T27 70 SRSWOR vs. SRSWR SRSWOR (n = 5) T4 16 T6 20 T10 22 T11 25 T26 32 SRSWR (n = 5) T7 20 T7 20 T20 35 T21 35 T25 46 Clustering sampling: Initial clusters T1 13 T2 15 T3 16 T4 16 T5 19 T6 20 T7 20 T8 21 T9 22 T10 22 T11 25 T12 25 T13 25 T14 25 T15 30 T16 33 T17 33 T18 33 T19 33 T20 35 T21 35 T22 36 T23 40 T24 45 T25 46 T26 52 T27 70 Cluster sampling (m = 2) T6 20 T7 20 T8 21 T9 22 T10 22 T21 35 T22 36 T23 40 T24 45 T25 46 Stratified Sampling T1 13 young T2 15 young T3 16 young T4 16 young T5 19 young T6 20 young T7 20 young T8 21 young T9 22 young T10 22 young T11 25 young T12 25 young T13 25 young T14 25 young T15 30 middle age T16 33 middle age T17 33 middle age T18 33 middle age T19 33 middle age T20 35 middle age T21 35 middle age T22 36 middle age T23 40 middle age T24 45 middle age T25 46 middle age T26 52 middle age T27 70 senior Stratified Sampling (according to age) T4 16 young T12 25 young T17 33 middle age T25 46 middle age T27 70 senior Figure 2.7: Examples of sampling: SRSWOR, SRSWR, cluster sampling, strati¯ed sampling. Based on the Euclidean distance of the normalized points, the order is x1; x3; x4; x2; x5, which is the same as the cosine similarity order. 2.18. ChiMerge [Ker92] is a supervised, bottom-up (i.e., merge-based) data discretization method. It relies on Â2 analysis: adjacent intervals with the least Â2 values are merged together till the chosen stopping criterion satis¯es. (a) Brie°y describe how ChiMerge works. (b) Take the IRIS data set, obtained from Machine Learning Data Repository), as a data set to be discretized. Perform data discretization for each of the four numerical attributes using the ChiMerge method. (Let the stopping criteria be: max- interval = 6). You need to write a small program to do this to avoid clumsy numerical computation. Submit your simple analysis and your test results: split points, ¯nal intervals, and your documented source program. Answer: (a) Brie°y describe how ChiMerge works. The basic algorithm of chiMerge is: begin sort values in ascending order 26 CHAPTER 2. DATA PREPROCESSING assign a separate interval to each distinct value while stopping criteria not met begin compute Â2 of every pair of adjacent intervals merge the two intervals with smallest Â2 value end end (b) Take the IRIS data set, obtained from Machine Learning Data Repository), as a data set to be discretized. Perform data discretization for each of the four numerical attributes using the ChiMerge method. (Let the stopping criteria be: max- interval = 6). You need to write a small program to do this to avoid clumsy numerical computation. Submit your simple analysis and your test results: split points, ¯nal intervals, and your documented source program. The ¯nal intervals are: Sepal length: [4.3 - 4.8],[4.9 - 4.9], [5.0 - 5.4], [5.5 - 5.7], [5.8 - 7.0], [7.1 - 7.9]. Sepal width: [2.0 - 2.2], [2.3 - 2.4], [2.5 - 2.8], [2.9 - 2.9], [3.0 - 3.3], [3.4 - 4.4]. Petal length: [1.0 -