SA – REVISED V1
FUNDAMENTAL CONCEPTS
- Data driven decision making (DDD) = practice of basing decisions on the analysis of data
o The more data-driven a firm is the more productive it is, even controlling for a wide range of
possible confounding factors
o Correlated with higher ROA, ROE, asset utilization and market value, relation seems to be causal
- Data science = understanding phenomena through the analysis of data à collection, storage, analysis and
implementation
o Needs access to data
o Supports DD but is also overlapping
o Critical skill is the ability to decompose a problem into pieces such that each piece matches a
known task for which tools are available
- Big data = very large datasets with 3 characteristics: volume, variety and velocity (data self has no
meaning)
- Data mining = extracting knowledge from data via technologies
- Data analysis = process of examining datasets to draw conclusions about the useful information
- Sort decision of interest = decisions which need discovery within data and are repeated on a large scale
TYPES OF DATA ANALYSIS
- Descriptive = what has happened? à simple descriptive statistics, dashboards
- Predictive = what could happen? à segmentation, regression etc
- Prescriptive = what should we do? à complex models for stock, production and logistics
- How can this lead to a strategic asset? à data and the capability to extract useful knowledge from data can
be a strategic asset
METHODS FOR DATA ANALYSIS
- Classification = assign each instance to predefined category
- Scoring = estimates the probability that an instance belongs to a specific class
- Regression = predicts a numerical outcome for each instance
- Similarity matching = find items that are most similar based on attributes
- Clustering = groups based in similarity, without target variable
- Co-occurrence = identifies events or items that frequently occur together
- Profiling = describes the typical behavior or characteristics of a group
- Link prediction = predicts future connection between entities
- Data reduction = compress many variables into fewer meaningful dimensions
- Causal modeling = identifies cause-effect relationships to determine impact of data
DATA MINING PROCES - CRISP
- Business understanding = understand problem, what do we want to do
- Data understanding = understand strengths and weaknesses of the data
- Data preparation = analytical technologies and converting the data (cleaning etc)
- Modeling = data mining techniques + output is model or pattern
, - Evaluation = assessment of results to business goals
- Deployment = putting results into real use, recoding new system to speed and compatibility
LEARNING MODELS
- Supervised = has specific target variable à used for classification, regression and causal modeling, to
predict, predictive
- Unsupervised = has NO specific target variable = used for clustering, co-occurrence and grouping, to
identify, descriptive
TERMINOLOGY
- Dataset = sample, population, data, set, workset
- Entity = object, instance, observation, element, example, line, row, feature, vector
- Attribute = feature, characteristic, variable, column
- Model = a simplified representation of reality created to serve a purpose
o Unsupervised = to identify à descriptive
o Supervised = to predict à predictive
- Induction = data à theory = developing classification + regression models, first sample then model
- Deduction = theory à data = using classification and regression to test hypothesis, applying model to
situation
- Entropy = measures how mixed/impure a segment of a dataset is
o Lower = purer
o Higher = more information
- Information gain (IG) = builds on theory and helps us determine which features is the most informative
o High IG = more information separation in the data
o Go from parent set (full set) to children set (splits)
CLASSIFICATION TREES
- Predictive model that uses a sequence of binary splits based on highest IG to classify observations into
pure, homogeneous groups à stop splitting when node is pure (homogeneous)
- Structural mapping of binary decisions that lead to a decision about the class of an object
LOGISTIC REGRESSION
- Predicts the probability of x belonging to class c using s-shaped function à models’ output is directly a
probability between 0 and 1
o Used for class probability discrimination
§ Pro
• Importance of individual factors
• Pretty well-known
§ Cons
• Time-consuming, no solution, minimum observations
LAPLACE CORRECTION
- Used to adjust probability estimates when working with small samples where raw probabilities can be
misleading or optimistic and include bias
FUNDAMENTAL CONCEPTS
- Data driven decision making (DDD) = practice of basing decisions on the analysis of data
o The more data-driven a firm is the more productive it is, even controlling for a wide range of
possible confounding factors
o Correlated with higher ROA, ROE, asset utilization and market value, relation seems to be causal
- Data science = understanding phenomena through the analysis of data à collection, storage, analysis and
implementation
o Needs access to data
o Supports DD but is also overlapping
o Critical skill is the ability to decompose a problem into pieces such that each piece matches a
known task for which tools are available
- Big data = very large datasets with 3 characteristics: volume, variety and velocity (data self has no
meaning)
- Data mining = extracting knowledge from data via technologies
- Data analysis = process of examining datasets to draw conclusions about the useful information
- Sort decision of interest = decisions which need discovery within data and are repeated on a large scale
TYPES OF DATA ANALYSIS
- Descriptive = what has happened? à simple descriptive statistics, dashboards
- Predictive = what could happen? à segmentation, regression etc
- Prescriptive = what should we do? à complex models for stock, production and logistics
- How can this lead to a strategic asset? à data and the capability to extract useful knowledge from data can
be a strategic asset
METHODS FOR DATA ANALYSIS
- Classification = assign each instance to predefined category
- Scoring = estimates the probability that an instance belongs to a specific class
- Regression = predicts a numerical outcome for each instance
- Similarity matching = find items that are most similar based on attributes
- Clustering = groups based in similarity, without target variable
- Co-occurrence = identifies events or items that frequently occur together
- Profiling = describes the typical behavior or characteristics of a group
- Link prediction = predicts future connection between entities
- Data reduction = compress many variables into fewer meaningful dimensions
- Causal modeling = identifies cause-effect relationships to determine impact of data
DATA MINING PROCES - CRISP
- Business understanding = understand problem, what do we want to do
- Data understanding = understand strengths and weaknesses of the data
- Data preparation = analytical technologies and converting the data (cleaning etc)
- Modeling = data mining techniques + output is model or pattern
, - Evaluation = assessment of results to business goals
- Deployment = putting results into real use, recoding new system to speed and compatibility
LEARNING MODELS
- Supervised = has specific target variable à used for classification, regression and causal modeling, to
predict, predictive
- Unsupervised = has NO specific target variable = used for clustering, co-occurrence and grouping, to
identify, descriptive
TERMINOLOGY
- Dataset = sample, population, data, set, workset
- Entity = object, instance, observation, element, example, line, row, feature, vector
- Attribute = feature, characteristic, variable, column
- Model = a simplified representation of reality created to serve a purpose
o Unsupervised = to identify à descriptive
o Supervised = to predict à predictive
- Induction = data à theory = developing classification + regression models, first sample then model
- Deduction = theory à data = using classification and regression to test hypothesis, applying model to
situation
- Entropy = measures how mixed/impure a segment of a dataset is
o Lower = purer
o Higher = more information
- Information gain (IG) = builds on theory and helps us determine which features is the most informative
o High IG = more information separation in the data
o Go from parent set (full set) to children set (splits)
CLASSIFICATION TREES
- Predictive model that uses a sequence of binary splits based on highest IG to classify observations into
pure, homogeneous groups à stop splitting when node is pure (homogeneous)
- Structural mapping of binary decisions that lead to a decision about the class of an object
LOGISTIC REGRESSION
- Predicts the probability of x belonging to class c using s-shaped function à models’ output is directly a
probability between 0 and 1
o Used for class probability discrimination
§ Pro
• Importance of individual factors
• Pretty well-known
§ Cons
• Time-consuming, no solution, minimum observations
LAPLACE CORRECTION
- Used to adjust probability estimates when working with small samples where raw probabilities can be
misleading or optimistic and include bias
