Improving sentiment classification accuracy through information gain feature selection and hyperparameter tuning in random forest model

¹ Department of Informatics Engineering, Faculty of Engineering, Universitas Trunodjoyo Madura, Bangkalan, Indonesia
² Department of Sharia Economics, Faculty of Islamic Studies, Universitas Trunodjoyo Madura, Bangkalan, Indonesia

^* Corresponding author: doni.fatah@trunojoyo.ac.id

Published online: 22 December 2025

Abstract

This study aims to enhance the accuracy of sentiment classification for Madura coastal tourism reviews using Random Forests optimised via Information Gain (IG) feature selection and hyperparameter tuning. The research contribution is an efficient hybrid framework that improves classification performance by integrating data-driven feature-relevance analysis and parameter optimisation techniques. The methodology involves several stages, including data scraping from online travel platforms, text preprocessing (cleaning, tokenisation, and stemming), TF-IDF feature extraction, feature selection using IG thresholds, and model optimisation using GridSearchCV. Experimental results show that the combination of Information Gain and hyperparameter tuning significantly increases the Random Forest model’s accuracy. At an IG threshold of 0.0002, the model achieved the highest accuracy of 89.69%, while the 0.0004 threshold provided nearly similar accuracy (89.18%) with fewer features, making it more efficient. These findings indicate that careful feature selection and parameter tuning can improve model generalisation and efficiency. In conclusion, this study contributes to advancing sentiment classification in tourism data analytics through an optimised Random Forest model and provides insights for future research in data-driven modelling for decision support in the tourism sector.