Improving Software Reliability Through Predictive Fault Analysis Using Machine Learning Models

The dependability of software systems depends heavily on software reliability since predicting faults before release improves their operational properties. The research combines data preprocessing methods with Synthetic Minority Over-sampling Technique (SMOTE) to tackle the common issues in Software Fault Prediction (SFP) related to dataset class imbalance and computational complexity. A dataset from PROMISE Software Engineering Repository named CM1 served as the source for conducting this comparative analysis of ML models. The study included Decision Tree (DT), Random Forest (RF), Genetic Algorithm-Support Vector Machine (GA-SVM) and the newly proposed Light Gradient Boosting Machine (LightGBM). Performance evaluation demonstrated LightGBM excelled above all other classifiers in a test based on accuracy, precision, recall and F1 score by reaching 91% accuracy, 91% precision as well as 90% recall while achieving 92% F1 score. The research outcomes demonstrate how LightGBM effectively handles class imbalance and computational issues which result in better software reliability through early fault detection.