My Projects
Student Dropout Predictor
I built a Random Forest based student dropout risk system using behavioral, academic, and engagement data, achieving an ROC AUC of 0.86 (precision 78%, recall 82%, F1 0.80). SHAP analysis revealed attendance, assignment scores, and financial stability as the top risk drivers. A Python and SQLite pipeline then scores new data in real time and issues early-warning alerts.
Amazon Sales Trend Analysis
I built an end-to-end Amazon review analysis dashboard using Python, leveraging pandas for data ingestion and cleaning, NLTK’s VADER for sentiment scoring, and Matplotlib, Seaborn, and Plotly for interactive visualizations. The workflow processes line-delimited JSON review data, handles missing fields, converts timestamps, standardizes votes, computes sentiment and rating distributions, and exports standalone HTML dashboards for each product.
Automated Dermatology Data Analysis Pipeline
- Data Prep: Cleans and imputes a raw dermatology CSV, normalizes features, and splits into train/test sets.
- Modeling:
- Custom gradient-descent regression on age.
- Random Forest & k-NN classifiers with grid search, cross-validation, and stability testing.
- K-Means & DBSCAN clustering, tuned by ARI/silhouette and visualized with PCA.
- Results: Logs all metrics and parameters to JSON, generates publication-quality plots comparing models.
- Tech Stack: Python · NumPy · pandas · scikit-learn · Matplotlib
