Development of an Ensemble Machine Learning Model for Sentiment Analysis of University Students' Opinions on NELFUND

The Nigerian Education Loan Fund (NELFUND) was established to improve access to higher education by providing financial assistance to eligible students. Since its introduction, students have actively expressed their views on the programme through social media platforms and online discussion forums. Analysing these opinions is essential for understanding public perception and assessing the transparency and perceived benefits of the scheme. Traditional sentiment analysis methods often rely on individual machine learning algorithms whose predictive performance may be constrained by model-specific limitations. Ensemble learning provides an alternative by combining multiple classifiers to improve robustness and classification accuracy. This study develops and evaluates a voting-based ensemble framework for sentiment analysis of students' opinions on NELFUND. Opinions were collected from students of Emmanuel Alayande University of Education, Oyo, through WhatsApp discussion groups. The textual data were preprocessed using data cleaning, tokenisation, part-of-speech tagging, stop-word removal, stemming, and lemmatisation. Text features were generated using Term Frequency–Inverse Document Frequency (TF-IDF) and Count Vectorisation techniques. Five machine learning classifiers—Support Vector Machine, Multinomial Naïve Bayes, Logistic Regression, Random Forest, and Extreme Gradient Boosting (XGBoost)—were trained individually and subsequently combined using three ensemble strategies: Majority Voting, Weighted Average Voting, and Average Probability Voting. Experimental results indicate that the proposed models achieved strong sentiment classification performance for both the transparency and perceived benefits of NELFUND. Across the evaluated datasets, classification accuracies ranged from 92% to 94%, while the ensemble approaches demonstrated competitive performance relative to the individual classifiers. The findings further reveal that classifier effectiveness varies with the characteristics of the dataset, suggesting that no single model consistently outperforms others under all conditions. Overall, the study demonstrates that ensemble learning provides a reliable and practical approach for analysing students' opinions and offers valuable insights that can support the evaluation and continuous improvement of educational funding initiatives.