Explainable AI Framework for Multiclass University Recommendations Using SHAP and Counterfactual SHAP

This research aims to (1) develop a robust machine learning model for university recommendations and ranking using advanced ensemble learning techniques, specifically XGBoost, and (2) enhance model interpretability to support informed decision-making through SHAP (Shapley Additive Explanations) and Counterfactual SHAP. These methods improve transparency by explaining features that influence recommendations, ensuring alignment between student profiles and appropriate universities. The model development process involves five key steps: data collection from publicly available university admission data, data preparation and feature engineering (including encoding, feature scaling, and class weighting), testing various models for multiclass classification using GBT, XGBoost, LightGBM, CatBoost, Random Forest, and Multi-Layer Perceptron, performance evaluation using confusion matrices and ranking metrics, and hyperparameter tuning to optimize results. The final model achieved an accuracy of 88.69%, with a Mean Reciprocal Rank (MRR) of 94.74% and a normalized Discounted Cumulative Gain (nDCG) of 96.12% after hyperparameter tuning. By leveraging SHAP and Counterfactual SHAP, the model provides actionable insights into feature adjustments, thereby improving outcomes and supporting equitable and informed decision-making based on interpretable and reliable criteria.