TananiMouhsin/student-dropout-prediction-morocco

# Student Dropout Prediction — Morocco An AI-based early warning system that predicts school dropout risk at the student level, enabling timely intervention by school counselors. Built as part of a PFA internship application mini-challenge. ## Problem Morocco records nearly 295,000 school dropouts every year. The current system is reactive — students are only flagged after they have already left. This project builds a classifier that identifies at-risk students before dropout occurs, using academic, socioeconomic, and demographic features collected at enrollment and end of semester 1. ## Proposal See `proposal_dropout_prediction_morocco.pdf` for the full 2-page research proposal. ## Dataset Predict Students' Dropout and Academic Success — Martins et al. (2021), UCI ML Repository ID 697. Download the CSV and place it as `data/data.csv` before running the notebook. ## Results | Model | Accuracy | Recall | Precision | AUC | |---------------------|----------|--------|-----------|-------| | Logistic Regression | 0.869 | 0.803 | 0.792 | 0.919 | | Random Forest | 0.880 | 0.785 | 0.832 | 0.926 | | **XGBoost** | **0.885**| **0.806** | **0.830** | **0.933** | Best model selected by **Recall** — missing a dropout student carries a higher cost than a false alarm. ## Usage ```bash pip install pandas numpy matplotlib seaborn scikit-learn xgboost imbalanced-learn joblib ``` Open `dropout_prediction.ipynb` and run all cells top to bottom. The trained model will be saved to `models/best_model_XGBoost.pkl`. ## Author **Mouhsin Tanani** — Data Science, Big Data & AI Engineering Student ENSIASD, Taroudant, Morocco LinkedIn · GitHub