The evaluation of heuristic optimization techniques on text categorization with conventional machine learning algorithms and deep learning methodologies

This research proposes a new approach called MBO-NB for tackling feature selection challenges in high-dimensional text classification tasks. MBO-NB combines Migrating Birds Optimization (MBO) via Naive Bayes as a base classifier. To prioritize computational efficiency, we employ a preprocessing step using the Information Gain (IG) algorithm. This strategic approach significantly reduces the average number of features from 62,221 to 2,089, streamlining the subsequent analysis. Our experiments demonstrate that MBO-NB achieves superior feature reduction effectiveness compared to existing techniques. Notably, this reduction in features translates to boosted categorization success. The unification of Naive Bayes and MBO proves to be a successful strategy, resulting in a well-balanced solution. Furthermore, head-to-head contrast via Particle Swarm Optimization (PSO) reveal that MBO-NB performs better than PSO by a mean of 6.9% across four different experimental settings. This thesis presents meaningful insights for improving attribute selection models in text classification, paving the way for a more scalable and effective approach.