Enhancing corporate bankruptcy prediction and customer relationship sustainability through CNN-based financial feature transformation

Based on the comprehensive analysis of a dataset comprising 43,405 companies (with only 2,091 labeled as bankrupt), this study demonstrates that addressing severe class imbalance through appropriate resampling techniques significantly enhances bankruptcy prediction performance and, consequently, enables more reliable assessment of companies’ capacity to maintain sustainable customer relationships. Two balanced datasets were created: a downsampled dataset with 4,182 samples and a SMOTE-enhanced dataset with 82,628 samples, where the analysis revealed that company size (logarithm of total assets) and working capital emerged as the most discriminative financial indicators directly impacting customer service delivery capabilities and…