Aplicación de Técnicas de Machine Learning para la Predicción Temprana de Defectos en Proyectos de Software
TL;DR
The results show that ensemble learning models and deep architectures achieve superior predictive capacity, particularly in scenarios with high dimensionality and class imbalance, and confirm the strategic value of Machine Learning as a systematic tool for contemporary software engineering.
Abstract
This article examines the application of Machine Learning techniques for the early prediction of defects in software projects, aiming to enhance product quality and optimize development processes. The main objective is to evaluate the effectiveness of supervised algorithms and deep learning models in identifying defect‑prone components before deployment, thereby reducing the costs associated with late corrections. The methodological approach follows a quantitative strategy based on the collection and preprocessing of code metrics, change history, and process attributes, using validated datasets such as NASA, PROMISE, and Jureczko. Models including Random Forest, Support Vector Machines, Gradient Boosting, and deep neural networks were implemented and assessed through performance metrics such as F1‑score, AUC, and MCC. The results show that ensemble learning models and deep architectures achieve superior predictive capacity, particularly in scenarios with high dimensionality and class imbalance. Furthermore, the early integration of predictions into the development cycle supports the prioritization of code reviews, strengthens risk management, and improves technical decision‑making. These findings confirm the strategic value of Machine Learning as a systematic tool for contemporary software engineering.