Optimizing software fault prediction using decision tree regression and soft computing techniques

This research aims to develop a framework for software fault prediction (SFP) using machine learning techniques. A software fault may be the reason behind the failure of software functioning, and even a minor fault could cause the failure. Efficient SFP improves the overall quality and performance of the software products while streamlining the development process. The framework aims to reduce the cost and time involved in software development while optimizing the reliability of the software. It facilitates quick and efficient testing by identifying the modules that are likely to fail at the early stages of the project. Soft computing techniques provide an easy and effective solution for prediction problems. This study emphasizes the significance of soft computing approaches in SFP and highlights their role in improving computational efficiency, reducing development costs, and enhancing the reliability of software applications. Soft computing-based technique was proposed to address the prediction challenges. A metric suite was suggested, which includes a requirement-based metric and an adoption metric, designed by integrating process metrics of software development phases for fault prediction. It also designs decision tree regression (DTR)-based SFP model that uses these metrics as input and delivers predicted faults as output. The literature review reveals that only a few existing frameworks meet the requirement of implementing SFP models using a broad range of soft computing approaches for the same dataset. The suggested metric suite is validated by computing performance measures such as the area under curve (AUC), F-measure, precision, recall, and accuracy. The high-performance values of the suggested metric suite demonstrate its efficient fault prediction capability. The study also compares the performance of the suggested model with other adaptive neuro fuzzy inference systems (ANFIS), fuzzy-inference systems, and Bayesian-net-based SFP models, measured by root mean square error (RMSE), normalized root mean square error (NRMSE), the mean magnitude of relative error (MMRE), the balanced mean magnitude of relative error (BMMRE), and R-Squared. The suggested model outperforms others, achieving RMSE, MMRE, and R-Squared values of 3.54, 2.04 e-05, and 99.78, respectively. This study presents a highly efficient DTR based SFP model with more fault prediction accuracy than the existing SFP models. Implementation of this model is to significantly reduce costs and improve the time and effort of software development, making it an invaluable tool for software engineers.