Improving Sentiment Classification using Ensemble Learning

Sherwan A Abdullah (College of Education, University of Zakho, Duhok, Iraq)
Mohammed I Salih (Computer Information System Department, Duhok Polytechnic University, Iraq)
Omar M Ahmed (Computer Information System Department, Duhok Polytechnic University, Iraq)

International Journal of Informatics, Information System and Computer Engineering (INJIISCOM)
Vol. 6 No. 2 (2025): INJIISCOM: VOLUME 6, ISSUE 2, DECEMBER 2025

This study proposes an ensemble learning approach to enhance sentiment classification accuracy on the IMDB movie reviews dataset. We combined three diverse models: Logistic Regression, Random Forest, and a Bidirectional Long Short-Term Memory (LSTM) neural network. Text data was vectorized using bigram term frequency, optimizing it for traditional classifiers, while the LSTM captured sequential dependencies via an embedding layer. By aggregating predictions through majority voting, the ensemble preserves the interpretability of traditional models while leveraging the deep learning capabilities of the neural network. Experimental results demonstrate that our proposed ensemble method achieves an accuracy of 89.2%, outperforming the individual models. This highlights the effectiveness of integrating traditional machine learning with deep learning for robust sentiment analysis.