Strengthening Support Vector Machine Capability through Harris Hawks Optimization for Arabic Text Categorization

Abstract

The automated categorization of Arabic text remains a persistently difficult problem, largely because the language combines a productive root-and-pattern morphology, pervasive orthographic variation, and a scarcity of large, carefully annotated corpora—factors that together inflate the feature space and obscure the signal available to a learner. Within this setting, the Support Vector Machine (SVM) continues to serve as a strong discriminative baseline, yet its effectiveness hinges critically on how well its penalty term, kernel family, and kernel coefficient are configured, and on the dimensionality of the representation it must operate over. This study introduces a hybrid framework, termed SVM-HHO, in which Harris Hawks Optimization (HHO), a swarm-intelligence metaheuristic modelled on the cooperative besiege-and-pounce behavior of Harris' hawks—is entrusted with a single, unified search that simultaneously calibrates the SVM hyperparameters and isolates a compact, highly discriminative subset of Term Frequency–Inverse Document Frequency (TF-IDF) features. The framework was assessed on a balanced, six-domain corpus of 6,000 Modern Standard Arabic news articles spanning Sports, Politics, Economy, Health, Technology, and Culture. Under this evaluation, SVM-HHO attained an accuracy of 99.5% together with precision, recall, and F1-score of 99.4%, surpassing by clear margins the untuned baselines examined—standard SVM (92.3%), Random Forest (94.2%), K-Nearest Neighbors (90.5%), and Naïve Bayes (88.3%)—while its accuracy also stands favourably beside a previously published SVM-PSO model (99.2% accuracy) reported on a related multi-domain Arabic corpus. Taken together, these findings indicate that the adaptive interplay between exploration and exploitation that characterizes HHO yields a particularly well-matched joint configuration of parameters and features, producing an accurate yet computationally frugal classifier for Arabic Natural Language Processing.