Analytics of ensemble learning-based methods for Arabic text classification

Abstract

In recent years, the volume of unstructured documents has grown exponentially, necessitating a profound understanding of Machine Learning (ML) methods to accurately classify documents in various domains, including business, media, sports, and more. This study proposes the utilization of ensemble learning methods for the automated classification of new documents. Specifically, ML algorithms and their bagging, boosting, stacking, and voting methods are employed. A comparative analysis of these algorithms is conducted using three benchmark Arabic datasets. The experimental results demonstrate that ensemble learning methods outperform individual learners in terms of accuracy, precision, recall, and F-measure. Among the ensemble learning classifiers, the stacking and voting methods exhibit the most promising performance, achieving the highest accuracy of 95.20% with the ANTCorpus dataset. Additionally, voting has led to improved accuracy of 93.24% for the Khaleej-2004 dataset and 92.15% for the Watan-2004 dataset.