Validación de modelos de aprendizaje automático para la predicción de fallas de software en equipos informáticos, con visualización interactiva a través de un dashboard

Abstract

Abstract: he growing organizational dependence on computer systems has increased the need to adopt predictive approaches that enable the anticipation of software failures and the reduction of operational disruptions. The present study aims to validate machine learning models focused on predicting software failures in computer systems, integrating the results into an interactive dashboard to support decision-making processes. This research follows a quantitative, applied, and experimental approach and is structured in accordance with the CRISP-DM methodology. For the analysis, the public BGL (Blue Gene/L) dataset was used, which consists of authentic logs from high-performance computing systems. From this dataset, structural and temporal variables were derived for supervised modeling purposes. Several algorithms were evaluated, including Logistic Regression, K-Nearest Neighbors (KNN), Support Vector Machines (SVM), Random Forest, and HistGradientBoosting. Cross-validation techniques were implemented, and appropriate metrics were employed to address class imbalance, such as precision, recall, F1-score, and ROC-AUC. The results indicate that ensemble-based models achieve superior predictive performance. Overall, the study provides empirical evidence supporting predictive maintenance and strengthens the integration between advanced analytics and interactive visualization.