Scholarlink

<p>In financial crisis prediction, binary classifiers often aim to minimize the overall error rate. This approach can lead to the neglect of the minority class (crisis cases) in imbalanced datasets, resulting in a critical prediction bias. While XGBoost is a typical high-performance binary classifier, imbalanced financial data can cause its gradient-based learning to generalize poorly for the minority class. <b>To address this issue</b>,<b> this study aims to develop a novel Probabilistic Decision Space (PDS) optimization method that reduces algorithmic bias and improves the detection of financial crises.</b> The proposed PDS method works by rebalancing the classifier’s learning process. It strategically adjusts the decision threshold to better identify minority samples and incorporates a gradient enhancement mechanism to increase the model’s sensitivity to these critical cases. Furthermore, we employ the Simulated Annealing (SA) algorithm for effective feature selection, eliminating irrelevant financial features to enhance overall performance. Empirical evidence from Chinese listed companies shows that while standard classifiers achieve high precision by favoring majority samples, they do so with a high error rate for crises. In contrast, the PDS-SA-XGBoost model improves the detection rate of minority class samples by 12.85%. It achieves a peak precision of 93.55 and an AUC value of 94.23 four years prior to the crisis (T-4 period). The results demonstrate that the PDS framework effectively enhances the classification of minority samples, providing a valuable new approach for handling imbalanced data in financial applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Class-Imbalanced Financial Crisis Prediction Method Based on Probability Decision Space Optimization

  • 期刊论文

  • 发表日期：

    2026年2月17日

查看全文
  • Zengli Mao,
  • Xiaofang Chen,
  • Chong Wuc

摘要

In financial crisis prediction, binary classifiers often aim to minimize the overall error rate. This approach can lead to the neglect of the minority class (crisis cases) in imbalanced datasets, resulting in a critical prediction bias. While XGBoost is a typical high-performance binary classifier, imbalanced financial data can cause its gradient-based learning to generalize poorly for the minority class. To address this issue, this study aims to develop a novel Probabilistic Decision Space (PDS) optimization method that reduces algorithmic bias and improves the detection of financial crises. The proposed PDS method works by rebalancing the classifier’s learning process. It strategically adjusts the decision threshold to better identify minority samples and incorporates a gradient enhancement mechanism to increase the model’s sensitivity to these critical cases. Furthermore, we employ the Simulated Annealing (SA) algorithm for effective feature selection, eliminating irrelevant financial features to enhance overall performance. Empirical evidence from Chinese listed companies shows that while standard classifiers achieve high precision by favoring majority samples, they do so with a high error rate for crises. In contrast, the PDS-SA-XGBoost model improves the detection rate of minority class samples by 12.85%. It achieves a peak precision of 93.55 and an AUC value of 94.23 four years prior to the crisis (T-4 period). The results demonstrate that the PDS framework effectively enhances the classification of minority samples, providing a valuable new approach for handling imbalanced data in financial applications.