94etienne/ai_driven_field_recomendation_at_rwanda_polytechnic

An AI-powered Field Recommendation System for Rwanda Polytechnic. Includes full ML pipeline: preprocessing with no data leakage, training Neural Networks/XGBoost/Random Forest, cross-validation, feature importance, PCA, SHAP & LIME explainability, statistical tests, and a deployable recommendation engine. # 🎓 Field Recommendation System for Rwanda Polytechnic ### _AI-Powered Student Placement Using Machine Learning & Explainable AI (XAI)_ This repository contains a **complete end-to-end Machine Learning pipeline** for predicting the most suitable academic field for Rwanda Polytechnic (RP) students. The system is built using **Neural Networks, XGBoost, and Random Forest**, combined with **interpretability tools (SHAP & LIME)** and strict **no–data-leakage** preprocessing workflows. The model analyzes **TVET combinations, examination boards, and subject-level marks** to generate accurate, fair, and explainable field recommendations. --- ## 🚀 Key Features ### 🧠 Machine Learning Pipeline - Neural Network (Keras) - XGBoost - Random Forest - 5-Fold Stratified Cross-Validation - Overfitting & generalization analysis ### 📊 Feature Engineering & Analysis - One-Hot Encoding (train-only) - Standard Scaling (train-only) - Mutual Information - Recursive Feature Elimination (RFE) - Permutation Importance - PCA (Dimensionality Reduction) - Feature Perturbation (robustness test) ### 🔍 Explainability (XAI) - **SHAP values** for global feature impacts - **LIME** for local instance-based explanations - Subject–Field correlation analysis - Per-class performance insights ### 📈 Visualizations Included - ROC curves (multi-class) - Confusion matrices (raw & normalized) - Heatmaps (MI, correlation, perturbation) - PCA scree plots - CV score boxplots - Confidence distribution histogram - Feature importance charts ### 🎯 Deployable Recommendation System A reusable class: ```python FieldRecommendationSystem ``` Allows real-time predictions with: - Predicted field - Confidence level - Top-3 recommended fields - Support for unseen boards/combos --- ## 📁 Project Structure ``` . ├── ai_field_recommendation_at_rp.ipynb # Full ML pipeline and analysis ├── results/ │ ├── field_recommendation_model.h5 # Saved best model │ ├── field_recommendation_model.pkl │ ├─ …