Research on the Application of Moss Growth Optimization Algorithm in Machine Learning Hyperparameter Tuning: Taking Stroke Risk Assessment as an Example

Addressing the challenges of hyperparameter optimization in stroke risk models, this study proposes a tuning framework using the Moss Growth Optimization (MGO) algorithm. MGO mimics moss growth mechanisms—spore diffusion, dual reproduction, and cryptobiotic survival—to implement a three-phase strategy combining global exploration, local refinement, and adaptive escape. This enables efficient navigation of high-dimensional, non-convex hyperparameter spaces. Experiments on the Kaggle stroke dataset evaluated eight machine learning models. MGO-optimized XGBoost, AdaBoost, and GBDT achieved near-perfect classification: accuracy ≥99.98% and ROC-AUC=1.0000, outperforming traditional tuning methods. Linear discriminant analysis and KNN showed moderate performance, while simpler architectures (e.g., decision trees) suffered from overfitting. The framework enhances adaptability to complex medical data through balanced global-local search dynamics and improves minority-class recognition in imbalanced datasets. This work establishes a systematic approach for medical ML optimization, advancing technical foundations for precise clinical stroke risk assessment.