Predicting software defect quantities with machine learning methods

In the past, the software testing process was the most neglected of the processes in the software life cycle, but today it is seen as the most important criterion that determines the quality of the software. As software has become more complex in various sectors, the detection and prevention of software defects has become an increasingly difficult process. Traditional software testing approaches are often inadequate in uncovering all potential defects, leading to an increase in the cost of time and resources spent on debugging processes. In recent years, machine learning techniques have provided promising solutions for software defect prediction. In this thesis, different types of machine learning algorithms are evaluated on different datasets for software defect detection. A detailed literature review on software defect detection has been conducted. Since the studies on this subject in the literature have shown the best performances on NASA open datasets JM1, KC1, KC2, CM1, PC1, these datasets were used in this thesis. It was observed that the accuracy rates were high when "Support Vector Machines (SVM)", "Naive Bayes", "LightGBM", "XGboost", "Gradient Boosting", "Random Forest" machine learning methods were used to detect software defects. Within the scope of the study, it is aimed to predict software defects and to detect defects and defect quantities in advance and an overview of the use of machine learning methods for software defect detection is presented. The aim of this study is to investigate the effectiveness of machine learning methods in detecting software defects and to compare their performance with traditional defect detection techniques. The study uses a large dataset of software code and various metrics related to code quality, complexity and size. The accuracy distribution and error quantities of our models are calculated.