A Machine Learning Approach for Effective Software Defect Detection

This paper examines an efficient Machine Learning (ML) strategy for software defect detection, focusing on identifying defective and non-defective software files and investigating the possibility of using exclusively structural code metrics. The real-life dataset, from the source code of twenty-six N otepad++ releases generated with the LocMetrics tool, was used to classify files as defective or non-defective. Seven ML models, namely Decision Tree, Support Vector Machine, Logistic Regression, Naive Bayes, k-Nearest Neighbors, Random Forest, and Extra Trees, were applied and benchmarked using accuracy, precision, recall, F1-score and their macro/weighted metrics. All models were fine-tuned with Randomized Grid Search, trained using to-fold Cross Validation, and evaluated in two steps: test data evaluation and extra unseen data validation. Results revealed that the Extra Trees model outperformed others with an accuracy of 82.09% and retained its dominance in validation and evaluation procedures. Lastly, the most important predictors were the type of source codes (presence of C++), BLOC, and SLOC-L, illustrating that the programming language, the clarity, and readability of the code (denoted by the number of blank lines), and the logical source lines of code play crucial roles in determining the likelihood of software defects.