<p>This systematic review presents a comprehensive analysis of hybrid optimization frameworks that combine hyperparameter-tuning and feature selection to enhance machine learning (ML) models for disease classification. Unlike previous studies that treat these processes separately, this study demonstrates how their integration improves diagnostic accuracy, computational efficiency, and clinical interpretability. Hyperparameter-tuning—optimizing learning rates, regularization terms, and architectural parameters—ensures models adapt effectively to complex biomedical data. Meanwhile, feature selection techniques (filter, wrapper, and embedded methods) identify critical biomarkers, reducing dimensionality and mitigating overfitting risks. This review findings reveal that simultaneous optimization strategies, such as metaheuristic algorithms combined with ML, outperform sequential approaches, achieving 12–15% higher accuracy in classifying cardiovascular, cancer, diabetes, oncological, and metabolic disorders. This study evaluates key trade-offs between computational cost and model performance, emphasizing robust cross-validation and explainable AI design to facilitate clinical adoption. Case studies illustrate how Bayesian optimization with LASSO-based feature selection enhances cancer detection sensitivity, while grid search paired with correlation-based selection improves cardiovascular risk prediction. Practical recommendations include prioritizing metaheuristic-driven workflows for high-dimensional data and validating models across multi-centre cohorts to ensure generalizability. Future research should explore deep reinforcement learning for autonomous hyperparameter-tuning and federated feature selection to address data privacy constraints. By unifying these optimization paradigms, this review advances the development of clinically actionable ML tools, fostering timely diagnosis and personalized therapeutic strategies.</p>