The Effect of Grid Search-Based Hyperparameter Tuning on Multinomial Naïve Bayes Model Performance in Classifying Stress Levels Among University Students

Stress among university students is an increasingly prominent issue, given the high levels of academic, social, and emotional pressures they encounter. Early identification of stress indicators is essential to prevent the development of more severe mental health issues. This study investigates the effect of hyperparameter tuning using grid search on the performance of a Multinomial Naïve Bayes (MNB) model for classifying student stress levels. The dataset was collected via web scraping from social media platforms where students express their emotions and experiences. Text preprocessing and feature extraction were performed using the TFIDF method, and stress levels were categorized into three classes: No Stress (0), Mild Stress (1), and High Stress (2). The MNB model was trained and evaluated using k-fold crossvalidation, with performance assessed via accuracy, precision, recall, and F1-score. Results indicate that grid search-based hyperparameter tuning significantly improved classification performance, increasing accuracy from 72.3% to 79.6%. These findings highlight that even simple models like Naïve Bayes can benefit substantially from systematic hyperparameter optimization, particularly in the context of stress detection from student-generated text data.