Ensemble Learning Methods for Improving SMS Spam Classification Accuracy

Abstract

SMS spam classification remains a critical challenge due to the short length, informal structure, and high variability of text messages. The performance of single machine learning and deep learning models is promising; however, the class imbalance problem, overfitting issue, and poor generalization across datasets frequently occur. In this paper, we investigate how ensemble learning can enhance the peak performance and stability in SMS spam filtering. Several ensemble models (bagging, boosting, stacking and voting based) are examined using heterogeneous base learners such as Naïve Bayes, Support Vector Machines, Random Forests or neural network models. The experimental results on benchmark SMS spam datasets show that the ensemble models consistently achieve higher performance compared with single classifiers in different measures, including accuracy, precision, recall and F1-score. The results demonstrate the advantages of ensemble learning in learning complementary decision boundaries and mitigating model bias and variance. The present work demonstrates the promise of ensemble methods for the development of trustworthy, scalable SMS spam filters to be efficiently applied in a production environment.