Hierarchical cluster-based IELM for financial distress prediction with imbalanced data.

Financial distress (FD) occurs when external economic factors or internal financial challenges threaten the stability of a business, often leading to financial difficulties or bankruptcy. Predicting FD is critical, but existing methods suffer from limitations such as a narrow focus on input variables, inadequate emphasis on key financial indicators, and challenges in handling large, imbalanced datasets. Besides, most of the existing techniques often fail to minimize computational complexity while enhancing forecast accuracy, necessitating the development of more effective models. Recognizing these challenges, this work proposes the APCIELM model, a multistage approach specifically designed for FDP with imbalanced datasets. Davies–Bouldin index-based hierarchical K-means clustering approach is devised to group data samples, followed by a new strategic differentiation between minority and majority classes. The proposed Rotation Affinity Propagation Cluster-based hypothesis determines the necessity for oversampling within specific clusters based on data distribution characteristics. Finally, an incremental extreme learning machine (IELM) model is employed for FDP which optimizes computational efficiency by eliminating ineffective calculations while maintaining high prediction performance. The results demonstrate that the proposed multistage prediction model outperforms single-stage models when dealing with imbalanced data. The efficiency of APCIELM model is evaluated using different metrics, including accuracy, precision, recall, F-score, and time complexity. The comprehensive analysis reveals the superior performance of the APCIELM model over the existing methods. [ABSTRACT FROM AUTHOR]