Análisis predictivo y gestión optimizada de la cartera de clientes morosos en CNEL EP, Unidad de Negocio Bolívar

Payment delinquency represents a structural barrier to the financial sustainability of public electricity distribution companies; however, unlike the financial sector, predictive models based on machine learning have not yet been adapted to this context. In particular, CNEL EP faces challenges in anticipating customer defaults, which limits the effectiveness of its commercial management. This study aims to identify the most accurate machine learning model to predict delinquency risk in the Bolívar Business Unit. A methodological approach based on Design Science Research and CRISP-DM was adopted, incorporating a systematic literature review (PRISMA), the analysis of 72,483 historical records, and the application of techniques such as PCA, SMOTE, and ensemble models (RandomForest, Gradient Boosting, AdaBoost, and VotingClassifier). Gradient Boosting and VotingClassifier achieved near-perfect performance metrics (Accuracy: 0.9982; F1 Macro: 0.9957; AUC ROC: 1.000) and (Accuracy: 0.9983; F1 Macro: 0.9959; AUC ROC: 1.000), even under stress scenarios involving noise, imbalance, and data loss. Furthermore, the integration of SHAP and LIME enabled transparent interpretation of predictions for non-technical users. The findings show that the proposed solution is robust, replicable, and applicable in practice. This study makes a significant contribution by demonstrating that machine learning models can enhance portfolio management in Ecuador's public electricity sector.