Resumen
Software bugs and program defects have significant negative effect on the cost and duration of software development process. Finding such bugs in early stages of the development process will cuts development time and maintenance costs. This investigation presents three different machine learning algorithms: K-Nearest Neighbors (KNN), Random Forest (RF), and Multilayer Perceptron (MLP) to build a new proposed software defect prediction model using different types of software performance metrics. This proposed model was tested on three public datasets obtained from NASA to assess its accuracy and revealed that the KNN was outperforms RF and MLP.