A Comparative Study of Extreme Learning Machine Variants for Sentiment Analysis on Railway App Reviews
Nuki Pujiani Yosephine, Budi Warsito, Dinar Mutiara Kusumo Nugraheni · INTERNATIONAL JOURNAL OF MATHEMATICS AND COMPUTER RESEARCH · 2025
This study compares three Extreme Learning Machine (ELM) variants: ELM, Weighted-ELM (WELM), and Boosting Weighted-ELM (BWELM) for sentiment analysis of user reviews from Indonesian railway applications. Using TF-IDF for feature extraction and IndoBERT for labeling, the models were evaluated on an imbalanced dataset. ELM achieved 68.25% accuracy but struggled with minority classes. WELM improved performance to 72% by addressing class imbalance. BWELM, combining WELM and AdaBoost, achieved the best result with 78.25% accuracy, effectively handling imbalanced data. The findings highlight BWELM's potential for sentiment analysis in real-world, imbalanced datasets.