Fault Prediction in Android Systems through AI

Software code defect prediction is important in improving code quality and the turnaround time of software products. In this thesis we investigate how to create and extract features, analyze existing work to create and realize a defect prediction technique that can be applied in an industrial setting. We conduct this investigation on version controlled source code from Git and Jira data. We identify and define metrics to be collected and build four Machine Learning (ML) models to predict if a file is clean or defective. We create a Cost Effectiveness (CE)evaluation technique to measure the performance of our ML models and achieve a score of 87% and an accuracy of 88 % on our best models