Advanced metaheuristic-based machine learning models have emerged as powerful tools for predicting stock performance, leveraging their ability to handle complex patterns and optimize model parameters effectively. These models, employing algorithms such as genetic algorithms, particle swarm optimization, simulated annealing, and ant colony optimization, excel in capturing non-linear relationships and adapting to dynamic market conditions. They are particularly valuable in feature selection, model training, and parameter tuning tasks, enhancing predictive accuracy and mitigating overfitting risks inherent in traditional statistical methods. This review examines their application in stock prediction, discussing empirical studies and comparative analyses that demonstrate their performance across diverse market scenarios and asset classes. While these models offer significant advantages, challenges such as interpretability, computational complexity, and sensitivity to hyperparameters remain pertinent. Future research should focus on improving algorithm efficiency, integrating alternative data sources, and addressing ethical considerations in algorithmic trading.