Accurate diagnosis of chronic diseases from clinical datasets is often hindered by high-dimensional features, class imbalance, and complex inter-feature dependencies. We proposed a novel feature selection and ensemble learning framework for disease diagnosis based on Hybrid Swarm Intelligence (HSIE) to overcome these difficulties. Using a newly created multi-objective fitness function, our approach combines the advantages of the Firefly Algorithm (FA), Particle Swarm Optimization (PSO), and Differential Evolution (DE) to find the best subset of features. In order to guarantee predictive performance and model interpretability and support evidence-based clinical decision support, this function concurrently maximizes feature relevance, reduces redundancy, and encourages sparsity. To improve classification on imbalanced healthcare data, we introduce an adaptive sampling strategy that dynamically selects among Synthetic Minority Oversampling Technique (SMOTE), Borderline-SMOTE, Adaptive Synthetic Sampling (ADASYN), and SMOTE-Edited Nearest Neighbors (ENN) based on the dataset’s imbalance ratio. The selected features are used to train a diverse set of state-of-the-art classifiers, including XGBoost, LightGBM, CatBoost, and neural networks, combined through soft voting and stacked ensemble learning. Using a real-world thyroid clinical dataset, experiments show that our method produces better results in terms of accuracy, F1-score, and ROC-AUC while lowering model complexity and preserving computational efficiency appropriate for clinical settings. For automated disease prediction in clinical decision support systems, this framework offers a reliable, comprehensible, and expandable solution.