Performance Analysis of Logistic Regression for Multiclass Classification of the Impact of AI Use on Students

Abstract

This study aims to analyze the performance of the Logistic Regression model in multiclass classification of the impact of Artificial Intelligence (AI) use on college students. The dataset used comes from the Kaggle platform and consists of 1000 observations with three target categories: Positive, Neutral, and Negative. Two Logistic Regression approaches were compared: One-vs-Rest (OvR) and Multinomial Logistic Regression. Evaluation was conducted using accuracy, precision, recall, and F1-score metrics with an 80:20 data split scenario. The experimental results showed that both approaches produced identical performance with an accuracy of 94.43% and an average F1-score of 0.92. These findings indicate that on the dataset used, the OvR approach provides equivalent performance to the multinomial approach. However, this study did not use statistical significance tests or cross-validation, so the results are evaluative and limited to the analyzed dataset. This study emphasizes the importance of selecting a simple and interpretable model in AI-based educational data analysis.