Elite elimination osprey optimization algorithm optimized kernel extreme learning machine for bankruptcy prediction problems

This study addresses the limitations of traditional bankruptcy prediction models, which often struggle with nonlinear financial data due to their tendency to fall into local optima, low efficiency in parameter optimization, and insufficient prediction accuracy. To overcome these issues, we propose a novel bankruptcy prediction model, EEOOA-KELM, which integrates the Elite-Elimination Osprey Optimization Algorithm (EEOOA) with Kernel Extreme Learning Machine (KELM). First, the standard Osprey Optimization Algorithm is enhanced by incorporating three core mechanisms: an elite-guided Lévy mutation strategy, a precise elimination and generation mechanism, and a global-best-guided boundary control. These improvements effectively balance the algorithm’s global exploration and local exploitation capabilities. Benchmark evaluations on the CEC2020 and CEC2022 test suites demonstrate that EEOOA achieves superior convergence speed and solution accuracy on unimodal, multimodal, and hybrid test functions in both 10- and 20-dimensional settings, significantly outperforming seven state-of-the-art algorithms, including the Grey Wolf Optimizer and Whale Optimization Algorithm. Building upon this, EEOOA is employed to optimize the kernel parameters and regularization coefficient of KELM, using the classification error rate from 10-fold cross-validation as the fitness function, thus constructing the EEOOA-KELM bankruptcy prediction model. Experiments on the Wieslaw enterprise bankruptcy dataset reveal that the proposed model achieves higher performance across multiple metrics—accuracy (76.6677%), precision (74.6096%), recall (77.6439%), and F1-score (75.5041%)—compared with competing models, while also demonstrating better stability and generalization ability. Further analyses of the fitness iteration curves and boxplots confirm that EEOOA provides faster convergence and lower error rates when optimizing KELM parameters, with minimal performance fluctuations across multiple cross-validation runs. Overall, this research introduces an efficient and reliable method for financial risk early warning, offering both significant theoretical contributions and practical value.