Research on the occurrence prediction of solar energetic particle events based on GA-XGBoost model

<jats:p> <jats:bold>Abstract</jats:bold> &lt;p&gt;This work relies on multi-source observational data from the 1997-2025 solar activity cycles, extracting various solar activity features such as flares, coronal mass ejections, and radio bursts. It constructs a dataset containing approximately 2,000 solar eruption events, and classifies the samples into SEP event and NonSEP event categories to conduct prediction study. The XGBoost model is chosen as the base prediction model. To address the issues of local optima, over-fitting, and insufficient generalization ability when using traditional grid search for parameter tuning, the genetic algorithm(GA) is introduced to conduct global adaptive optimization for the core hyperparameters of the model. Experimental results show that after standard preprocessing and parameter optimization, this GA-XGBoost model can effectively mine the correlation between solar activity parameters and SEP events. Through SHAP explainability analysis, seven core influencing features are selected, and simplifying the features, the model still maintains a high prediction performance. Comparative experiments of multiple models confirm that compared with SVM, traditional XGBoost, and AdaBoost models, GA-XGBoost has significant advantages in accuracy, precision, recall rate, F1-score and various error evaluation indicators. It can effectively improve the prediction ability of SEP events, alleviate the prediction bias caused by the imbalance of sample categories, and provide efficient and feasible new ideas and technical support for space weather warning and SEP event prediction. </jats:p>