Predictive Modeling of Student Exam Performance Based on Learning Habits Using Machine Learning and SHAP Analysis

Abstract

This study aims to analyze the impact of Artificial Intelligence (AI) use on student academic performance using a machine learning approach with a Logistic Regression model for multiclass classification. The dataset used consists of 1000 observations obtained from the Kaggle platform and includes variables related to the intensity of AI use and student academic indicators. The research stages include data preprocessing, 80:20 data splitting, training the Logistic Regression model, and evaluation using accuracy, precision, recall, and F1-score. The results show that the model is capable of classifying with stable performance across classes. Interpretive analysis using SHAP was used to understand the contribution of each variable to the prediction results. The findings indicate that several factors of AI use have different influences on student academic performance categories, with relationship patterns that can be explained transparently using an Explainable Artificial Intelligence (XAI) approach. This research is evaluative and interpretative in nature, and does not claim generalizations beyond the context of the dataset used. The results are expected to form the basis for developing further analysis with additional validation and more complex modeling approaches.