Scholarlink

<p><?tk 2?>Corporate bankruptcy prediction is essential for assessing companies’ capacity to maintain sustainable customer relationships and service quality. This study proposes a novel CNN-based hybrid approach that transforms correlation-filtered financial features into 64 × 64 grayscale images, enabling reliable identification of firms at financial risk whose deteriorating conditions could compromise their ability to maintain quality customer service and sustain long-term business relationships. The research explicitly examines the linkage between financial health indicators and customer relationship sustainability by categorizing financial features based on their operational impact on service delivery, relationship management capabilities, and long-term customer commitment fulfillment. The methodology was evaluated on a comprehensive dataset of 43,405 Polish companies (2,091 bankrupt, 41,314 healthy) using two resampling strategies: random downsampling and Synthetic Minority Oversampling Technique (SMOTE). Following correlation-based feature selection that reduced multicollinearity by eliminating features with absolute correlation coefficients exceeding 0.8, retained financial features were normalized and transformed into spatial image representations. Six classification models were implemented: Deep Neural Network (DNN), Support Vector Machine (SVM), Random Forest (RF), Decision Tree (DT), Gradient Boosting (GB), and Logistic Regression (LR), alongside five CNN-hybrid variants, evaluated using 5-fold cross-validation. SMOTE-balanced datasets demonstrated superior performance across all models. Ensemble methods achieved exceptional accuracy, with Random Forest reaching 99.99% and Gradient Boosting 99.97%. The innovative CNN-SVM hybrid model attained 99.77% accuracy with perfect ROC-AUC (1.000), providing reliable indicators for assessing firms’ financial stability and their ability to invest in customer experience initiatives. Statistical analysis identified company size and working capital as the most discriminative financial indicators directly impacting customer service delivery capabilities. Customer-related metrics such as receivables turnover and collection period indicators emerged as critical predictors of relationship management effectiveness. The study contributes a novel spatial feature representation methodology enabling precise identification of companies whose financial deterioration could compromise customer service quality and relationship sustainability. These findings provide significant implications for stakeholders seeking enhanced risk assessment capabilities that consider both internal financial health and external customer relationship dynamics in bankruptcy prediction.</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

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

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月2日

查看全文
  • Ali Shahbazi,
  • Nefise Şirzad,
  • Sima Darvishan,
  • Amir Heidari Arash,
  • Zahra Fakhri,
  • Hossein Najafzadeh

摘要

Corporate bankruptcy prediction is essential for assessing companies’ capacity to maintain sustainable customer relationships and service quality. This study proposes a novel CNN-based hybrid approach that transforms correlation-filtered financial features into 64 × 64 grayscale images, enabling reliable identification of firms at financial risk whose deteriorating conditions could compromise their ability to maintain quality customer service and sustain long-term business relationships. The research explicitly examines the linkage between financial health indicators and customer relationship sustainability by categorizing financial features based on their operational impact on service delivery, relationship management capabilities, and long-term customer commitment fulfillment. The methodology was evaluated on a comprehensive dataset of 43,405 Polish companies (2,091 bankrupt, 41,314 healthy) using two resampling strategies: random downsampling and Synthetic Minority Oversampling Technique (SMOTE). Following correlation-based feature selection that reduced multicollinearity by eliminating features with absolute correlation coefficients exceeding 0.8, retained financial features were normalized and transformed into spatial image representations. Six classification models were implemented: Deep Neural Network (DNN), Support Vector Machine (SVM), Random Forest (RF), Decision Tree (DT), Gradient Boosting (GB), and Logistic Regression (LR), alongside five CNN-hybrid variants, evaluated using 5-fold cross-validation. SMOTE-balanced datasets demonstrated superior performance across all models. Ensemble methods achieved exceptional accuracy, with Random Forest reaching 99.99% and Gradient Boosting 99.97%. The innovative CNN-SVM hybrid model attained 99.77% accuracy with perfect ROC-AUC (1.000), providing reliable indicators for assessing firms’ financial stability and their ability to invest in customer experience initiatives. Statistical analysis identified company size and working capital as the most discriminative financial indicators directly impacting customer service delivery capabilities. Customer-related metrics such as receivables turnover and collection period indicators emerged as critical predictors of relationship management effectiveness. The study contributes a novel spatial feature representation methodology enabling precise identification of companies whose financial deterioration could compromise customer service quality and relationship sustainability. These findings provide significant implications for stakeholders seeking enhanced risk assessment capabilities that consider both internal financial health and external customer relationship dynamics in bankruptcy prediction.