Machine Learning Hyperparameters Optimization for Accurate Arabic Sentiment Classification

Abstract

An improved model performance is achieved by optimizing hyperparameters for Arabic sentiment classification based on machine learning. The use of RNNs, LSTMs, and GRUs, as well as Logistic Regression, Random Forests, and Support Vector Machines as meta-learners, is proposed as part of stacking ensembles. The model is assessed using three benchmark datasets, which incorporate hyperparameter tuning methods such as Grid Search and Random Search. Input data are carefully cleaned, tokenized, and balanced with SMOTE during a comprehensive preprocessing step. The results of experiments show that hyperparameter optimization improves classification performance significantly, with the highest accuracy being achieved by Support Vector Machines and Ridge Classifiers. Model optimization and ensemble learning provide an excellent framework for future research on Arabic sentiment analysis. Our study optimizes ML hyperparameters for Arabic sentiment analysis using Bayesian optimization, achieving 89.7% accuracy - a 6.2% improvement over baseline models. The method particularly enhances performance on dialectal Arabic content.