Machine Learning Course Final Project – Electrical Engineering Department ,Sharif University of Technology
| Part | Status | Summary |
|---|---|---|
| 1 – Classification | ✅ | Predicted at‑risk students (G3<10) across three temporal scenarios (T0, T1, T2). Evaluated Logistic Regression, Random Forest, XGBoost, SVM, KNN. Addressed class imbalance, feature importance, fairness across groups, and cost‑sensitive decision thresholds. Best model: XGBoost (F1=0.833, AUC=0.973 at T2). |
| 2 – Clustering & Regression | ✅ | Applied K‑Means, Hierarchical, Spectral, and DBSCAN to identify student groups (k=2 gives Weak/Strong clusters). Built regression models (XGBoost, RF, Linear, SVR, KNN) to predict final grades; best regressor: XGBoost (RMSE=0.261, R²=0.996). Engineered features, handled outliers, tuned hyperparameters, and compared regression‑to‑classification vs. direct classification. |
- Notebook:
Q1.ipynb - Covers 21 questions: risk identification, accuracy vs. F1, early warning reliability, effect of adding G1/G2, threshold changes, three‑level risk, excellent students, performance drops, recovery, limited‑capacity program selection, cost‑sensitive threshold, calibration, feature importance, error analysis, fairness, and cross‑subject generalizability.
- Notebook:
Q2.ipynb - Covers: selection of optimal k for clustering, comparison of clustering methods, regression model evaluation (bias‑variance), effect of removing G2, impact of engineered features, outlier handling, feature importance analysis, improvement ideas (GridSearchCV and ensemble), and conversion of regression to classification.
- Course: Machine Learning (25737)
- Instructor: Dr. Shamsollahi
- Department: Electrical Engineering, Sharif University of Technology
- Date: August 2026
- Mohammad Reza Mahdavi (401102637)
Project complete – all notebooks, data, and report are included.