Application of machine learning on process metrics for defect prediction in mobile application

Abstract

This paper studied process metrics in detail for predicting defects in an open source mobile applications in continuation with our previous study (Moser et al. Software Engineering, 2008). Advanced modeling techniques have been applied on a vast dataset of mobile applications for proving that process metrics are better predictor of defects than code metrics for mobile applications. Mean absolute error, Correlation Coefficient and root mean squared error are determined using different machine learning techniques. In each case it was concluded that process metrics as predictors are significantly better than code metrics as predictors for bug prediction. It is shown that process metrics based defect prediction models are better for mobile applications in all regression based techniques, machine learning techniques and neuro-fuzzy modelling. Therefore separate model has been created based on only process metrics with large dataset of mobile application.