Rambagira/ICGS_Master

A machine learning-based Intelligent Career Guidance System (ICGS) deploying a Random Forest classifier to recommend Rwandan secondary school (REB) and TVET pathways based on academic marks and vocational interests. # Intelligent Career Guidance System (ICGS) 🇷🇼 **Live Production Application:** icgs.onrender.com ## Overview The Intelligent Career Guidance System (ICGS) is a data-driven web application designed to optimize the transition of Rwandan students from Ordinary Level (O-Level) into Advanced Level (A-Level) combinations and TVET (Technical and Vocational Education and Training) streams. Developed as part of a Master's Thesis in Internet Systems at Kigali Independent University (ULK) by Iradukunda Jean Felix, this platform replaces manual, bias-prone counseling with an automated, highly scalable ensemble machine learning architecture. ## System Architecture & Features ### 1. Deterministic Policy Gatekeeper The system bridges probabilistic AI with strict national educational legislation. The Flask backend features a hardcoded regulatory filter that calculates the student's cumulative aggregate and blocks machine learning inference if the 50% REB (Rwanda Basic Education Board) promotion threshold is not met, seamlessly returning a remedial advisory notice instead. ### 2. The Machine Learning Engine (Random Forest) The analytical core utilizes a Scikit-Learn `RandomForestClassifier` trained on 8,000 synthetic student profiles generated via Archetype Clustering. * **Feature Matrix:** 19 columns (9 continuous numerical subject grades + 10 One-Hot Encoded categorical vocational interests). * **Target Classes:** 37 distinct educational pathways across General STEM, Humanities, Professional Tracks, and TVET Levels 3-5. * **Optimization:** To ensure serverless cloud stability and prevent overfitting, the model is strictly regularized to **50 decision trees** (`n_estimators=50`) and a `max_depth` of 20. * **Performance:** Utilizing SMOTE to correct class imbalances for elite tracks, the model achieves a highly realistic global cross-validated accuracy of **91.05%**. ### 3. Decoupled Cloud Deployment * **Frontend:** Responsive HTML5/CSS3/Bootstrap 5 interface o …