A two-step rough fuzzy clustering method for breaking predictions

Predicting corporate bankruptcy is a critical issue in the financial sector, requiring a comprehensive evaluation of a company's financial health to identify firms that may fail to meet their debt obligations. This issue not only has far-reaching implications for a company's development but also holds significant importance for socio-economic stability. The introduction of clustering algorithms into the field of enterprise risk management represents a significant advancement, providing a data-driven supplement and extension to traditional risk analysis methods. This paper introduces a novel two-step rough fuzzy clustering method (TRF-BPC) for bankruptcy prediction, leveraging machine learning algorithms to enhance traditional risk management practices and emphasizing the need for sophisticated evaluation strategies to address risks. Firstly, the TRF-BPC algorithm innovatively employs rough fuzzy sets to process financial data, constructs boundaries (optimistic and pessimistic predictions) based on the correlations between attributes, and performs clustering by comparing the similarity between objects and these boundaries. Secondly, to more effectively handle uncertainties, a divide-and-conquer strategy is adopted for two-stage refined clustering. Additionally, the algorithm optimizes partition thresholds through grid search, avoiding the limitations of fixed thresholds. The method first processes and groups the attribute set, then calculates prediction values for objects based on these attribute groups. Subsequently, objects are classified by comparing their similarity with the prediction values. The effectiveness of this algorithm has been validated on nine datasets, with additional analysis conducted on its noise resistance capability. This attribute-oriented clustering process enables the analysis of factors influencing corporate bankruptcy, explores the interrelationships between these factors, and enhances the authenticity and validity of the clustering process. By dynamically assessing risks through stepwise clustering, this method addresses shortcomings in traditional financial risk analysis, such as data dependency, normal distribution assumptions, and static analysis. The TRF-BPC algorithm, with its robustness, accuracy, and adaptability, has emerged as a powerful tool for bankruptcy prediction and risk management in the evolving financial landscape.