Improving Software Quality in Software Development using Random Forest Approach

In an era where software drives critical systems across industries, ensuring high-quality software before deployment is essential to enhance reliability, security, efficiency, and user satisfaction while minimizing risks, costs, and failures. One effective approach is defect management that prevent failures and enhance overall system performance. Datasets used for defect prediction suffer from class imbalance. Learning from imbalanced data and hyperparameter values of machine learning (ML) algorithms has affected the performance of prediction models that they generate. In this paper, we used JM1, one of the dataset from NASA promise dataset repository which suffer from class imbalance problem that leads to inaccurate classification and prediction. This research aims to improve the prediction of Random Forest (RF) classifier by addressing data imbalance. Specifically, it employs a hybrid approach combining Nearmiss undersampling and SMOTE oversampling to create a balanced dataset. The study then evaluates and compares F1-score of the enhanced RF model against Decision Tree (DT) and Adaptive Boosting (AdaBoost) classifiers. The experimental findings proved that Random Forest (RF) was the most effective classifier for detecting defects.