Software Defect Predictor and Classifier Tool Using Machine Learning Techniques

The Software testing is critical for ensuring the quality of software, but it can consume more than half of a software project's total expenses. This research can help software development teams identify potential bugs early on, leading to efficient and effective software creation. To improve the efficiency of software testing, this report presents research on software defect prediction using a hybrid machine learning algorithm that combines supervised and unsupervised learning techniques. The study used publicly available datasets to train and evaluate various machine learning models, including an ensemble learning model. A comparative analysis was conducted among KNN, Random Forest, Adaboost, Multilayer Perceptron, CNN, and Gaussian Naïve Bayes on different datasets. Feature selection was performed using various methods, including Recurrent Feature Elimination (RFE), SelectKBest, and Correlation, implemented by a tool created to select any feature selection algorithm and compare the results for various combinations with the above mentioned classifiers. The results achieved by the model are showed in detail for each of the algorithms on the selected datasets. The main objectives of the research are to create a tool to apply various machine learning techniques to the datasets and determine the most effective ones, as well as to explore different feature selection methods for improving the accuracy of defect prediction.