Explainable Ai-based Student Result Prediction Using Catboost And Logistic Regression

Abstract

Early Identification Of At-risk Students Can Significantly Improve Academic Outcomes. This Study Proposes An Explainable Artificial Intelligence (XAI)- Based Hybrid Model Combining Logistic Regression And CatBoost To Predict Student Performance. CatBoost Captures Complex Non-linear Relationships Among Factors Such As Attendance, Internal Marks, Study Hours, And Prior GPA, While Logistic Regression Provides An Interpretable Probability-based Decision Boundary. A Soft-Voting Ensemble Combines The Predictions Of Both Models, And SHAP (SHapley Additive ExPlanations) Is Used To Explain Feature Contributions. Using An 80:20 Train-test Split On 10,000 Student Records, The Proposed Model Achieved 99.10% Accuracy, 99.37% Precision, 99.02% Recall, 99.19% F1-score, And 99.98% ROC-AUC. SHAP Analysis Identified Study Hours, Prior GPA, And Internal Marks As The Most Influential Factors. The Proposed System Offers An Accurate, Interpretable, And Computationally Efficient Tool For Early Identification Of Academically At-risk Students

Keywords