A set of data mining projects built for CSCI 5523, each taking a dataset from exploratory analysis through model selection and evaluation.
Classification
Telecom customer churn was the primary dataset: distribution and correlation analysis first, then decision trees and k-nearest neighbors with hyperparameter tuning and cross-validation. A separate naive Bayes spam classifier covered text preprocessing, feature extraction, and precision/recall tradeoffs. The same pipeline was then run across the iris, diabetes, and thyroid datasets to compare model performance, using ROC curves for selection.
Unsupervised Learning and Association Mining
An Apriori implementation for market basket analysis, mining association rules by support and confidence over Instacart transaction data to surface purchase patterns. Cluster analysis covered k-means and hierarchical clustering, evaluated with silhouette scores and dendrograms. The final project clustered CORD-19 COVID-19 research papers by text embedding similarity to discover topic groupings without labels.
Stack
Python in Jupyter notebooks: pandas and NumPy for manipulation, scikit-learn for the algorithms, matplotlib and seaborn for visualization.