WESTERN GOVERNORS UNIVERSITY
D603 MACHINE
LEARNING
TASK 1: CLASSIFICATION DATA
MINING MODELS
Medical Readmission Prediction Using
Random Forest Classification
Program: M.S. Data Analytics
Assessment: D603 — Task 1
Data source: medical_clean.csv
Tools: Python, scikit-learn, pandas, Jupyter
Notebook, GitLab
2026–2027
, D603 | CLASSIFICATION DATA MINING
Contents
1. Executive Summary
2. Part A: GitLab Repository
3. Part B: Purpose and Goals
4. Part C: Classification Method
5. Part D: Data Preparation
6. Part E: Data Analysis and Results
7. Part F: Analysis Summary
8. Conclusion
9. References
10. Appendices: Code, Visualizations, and Submission Evidence
Evidence basis. The accompanying source report supplies the original dataset description, model
settings, and two confusion matrices. The numerical metrics shown here are calculated directly
from those matrices. The source report does not include the machine-readable dataset, executed
notebook, prediction probabilities, or final GitLab URL. Items requiring those source files are
specifically identified rather than presented as independently executed results.
Executive Summary
Hospital readmissions within 30 days create important challenges for patient
safety, care coordination, resource utilization, and healthcare quality
management. Patients who return unexpectedly after discharge may require
additional inpatient services, thereby increasing strain on available clinical
resources. The ability to identify increased readmission risk can help healthcare
organizations prioritize transitional-care services, medication reconciliation,
patient education, and follow-up planning.
This report evaluates a Random Forest classification approach using the
medical_clean.csv dataset, which the source analysis describes as containing
10,000 patient observations and 50 original columns. The target variable,
ReAdmis, indicates whether an observation is associated with a readmission
within approximately one month. Predictors include demographics, clinical
conditions, hospitalization characteristics, and patient-reported measures.
WESTERN GOVERNORS UNIVERSITY 2
, D603 | CLASSIFICATION DATA MINING
Random Forest was selected because it can model nonlinear patterns and
interactions among predictors while aggregating predictions from numerous
decision trees. The documented workflow consists of feature preparation,
stratified training/validation/test partitions, an initial classifier, five-fold cross-
validation for hyperparameter selection, and final classification evaluation.
The initial model’s documented training confusion matrix yields 98.47%
accuracy, 96.16% precision, 97.50% recall, and 96.82% F1 score. The
documented final evaluation matrix yields 98.50% accuracy, 96.30% precision,
97.51% recall, and 96.90% F1 score. These results exceed the project accuracy
objective of 85% within the recorded matrices. The source also reports ROC-
AUC values of 0.9992 and 0.9993; these cannot be calculated independently
without prediction probabilities.
One dataset reconciliation item remains: the source defines a 2,000-record test
partition, while the published final confusion matrix totals 2,002 records.
Therefore, the final metrics are clearly labeled as calculations from that
recorded matrix pending verification against actual test predictions. The
substantive interpretation emphasizes reproducibility, avoidance of data
leakage, clinical relevance, temporal generalizability, and careful
implementation.
Part A: GitLab Repository
A.1 Repository Creation and Management
GitLab serves as the centralized version-control environment for the
classification data mining project. It supports traceability of code changes,
documentation of analytical decisions, storage of generated output files, and
inspection of the development process. The repository should preserve a
sequence of commits corresponding to data import, preparation, initial
modeling, cross-validation, and final evaluation.
Repository structure and deliverables
Project component Function
README.md Project objective, requirements, reproducibility
instructions, and files
medical_readmission_analysis.ipynb Executable data preparation, training, and
evaluation workflow
medical_cleaned_final.csv Selected original features and encoded outcome
training_dataset.csv Stratified training partition
WESTERN GOVERNORS UNIVERSITY 3
D603 MACHINE
LEARNING
TASK 1: CLASSIFICATION DATA
MINING MODELS
Medical Readmission Prediction Using
Random Forest Classification
Program: M.S. Data Analytics
Assessment: D603 — Task 1
Data source: medical_clean.csv
Tools: Python, scikit-learn, pandas, Jupyter
Notebook, GitLab
2026–2027
, D603 | CLASSIFICATION DATA MINING
Contents
1. Executive Summary
2. Part A: GitLab Repository
3. Part B: Purpose and Goals
4. Part C: Classification Method
5. Part D: Data Preparation
6. Part E: Data Analysis and Results
7. Part F: Analysis Summary
8. Conclusion
9. References
10. Appendices: Code, Visualizations, and Submission Evidence
Evidence basis. The accompanying source report supplies the original dataset description, model
settings, and two confusion matrices. The numerical metrics shown here are calculated directly
from those matrices. The source report does not include the machine-readable dataset, executed
notebook, prediction probabilities, or final GitLab URL. Items requiring those source files are
specifically identified rather than presented as independently executed results.
Executive Summary
Hospital readmissions within 30 days create important challenges for patient
safety, care coordination, resource utilization, and healthcare quality
management. Patients who return unexpectedly after discharge may require
additional inpatient services, thereby increasing strain on available clinical
resources. The ability to identify increased readmission risk can help healthcare
organizations prioritize transitional-care services, medication reconciliation,
patient education, and follow-up planning.
This report evaluates a Random Forest classification approach using the
medical_clean.csv dataset, which the source analysis describes as containing
10,000 patient observations and 50 original columns. The target variable,
ReAdmis, indicates whether an observation is associated with a readmission
within approximately one month. Predictors include demographics, clinical
conditions, hospitalization characteristics, and patient-reported measures.
WESTERN GOVERNORS UNIVERSITY 2
, D603 | CLASSIFICATION DATA MINING
Random Forest was selected because it can model nonlinear patterns and
interactions among predictors while aggregating predictions from numerous
decision trees. The documented workflow consists of feature preparation,
stratified training/validation/test partitions, an initial classifier, five-fold cross-
validation for hyperparameter selection, and final classification evaluation.
The initial model’s documented training confusion matrix yields 98.47%
accuracy, 96.16% precision, 97.50% recall, and 96.82% F1 score. The
documented final evaluation matrix yields 98.50% accuracy, 96.30% precision,
97.51% recall, and 96.90% F1 score. These results exceed the project accuracy
objective of 85% within the recorded matrices. The source also reports ROC-
AUC values of 0.9992 and 0.9993; these cannot be calculated independently
without prediction probabilities.
One dataset reconciliation item remains: the source defines a 2,000-record test
partition, while the published final confusion matrix totals 2,002 records.
Therefore, the final metrics are clearly labeled as calculations from that
recorded matrix pending verification against actual test predictions. The
substantive interpretation emphasizes reproducibility, avoidance of data
leakage, clinical relevance, temporal generalizability, and careful
implementation.
Part A: GitLab Repository
A.1 Repository Creation and Management
GitLab serves as the centralized version-control environment for the
classification data mining project. It supports traceability of code changes,
documentation of analytical decisions, storage of generated output files, and
inspection of the development process. The repository should preserve a
sequence of commits corresponding to data import, preparation, initial
modeling, cross-validation, and final evaluation.
Repository structure and deliverables
Project component Function
README.md Project objective, requirements, reproducibility
instructions, and files
medical_readmission_analysis.ipynb Executable data preparation, training, and
evaluation workflow
medical_cleaned_final.csv Selected original features and encoded outcome
training_dataset.csv Stratified training partition
WESTERN GOVERNORS UNIVERSITY 3
