One of the most active areas of research in the software engineering community is defect prediction. The gap between data mining and software engineering must be bridged to increase the rate of software success. Before the testing phase, software defect prediction predicts where these flaws will occur in the source code. Methods for predicting software defects are widely used to investigate the impact area in software using various techniques (clustering, statistical methods, neural networks, and machine learning models). The goal of this research is to examine various machine learning algorithms for predicting software defects. There have been many fault prediction techniques introduced, but no single technique or approach can be used for all types of datasets. To achieve maximum accuracy, different machine learning algorithms such as Bayesian Net, Logistic Regression, Multilayer Perceptron, Ruler Zero-R, J48, Lazy IBK, Support Vector Machine, Neural Networks, Random Forest, and Decision stump were used to uncover the largest subset of defects that could be predicted. This research concern is to find out defects using five NASA data sets JM1, CM1, KC1, KC2, and PC1. Logistic Regression has been shown to have the best output compared to others (93%).