Automatic Parameter-Optimized XGBoost for Risk Early Warning Algorithm Application in Securities

Abstract

The stability of the securities market is crucial to a nation’s economy, and accurately predicting stock price crashes helps regulators implement protective measures in advance. Existing research has demonstrated that machine learning methods can accurately predict securities risks. Among them, the XGBoost algorithm, due to its characteristics of integrated learning, can combine the prediction results of multiple gradient boosting decision trees, thereby reducing the probability of false positives of the model. Therefore, this paper proposes a securities risk prediction model that integrates XGBoost with an evolutionary algorithm to enhance the accuracy and stability of stock crash prediction. To address the varying adaptability of different financial indicators in the model, an evolutionary algorithm is introduced to optimize XGBoost’s parameter configuration. Additionally, the classification task is transformed into a regression task, combined with a dynamic threshold-setting mechanism to mitigate data imbalance issues. Experimental results demonstrate that the proposed method outperforms baseline models across multiple datasets and maintains strong stability under different training-test set ratios, achieving an average performance improvement of 15.52% to 25.79% compared to state-of-the-art (SOTA) methods.