Software Engineering Approach to Bug Prediction Models using Machine Learning as a Service (MLaaS)

Abstract

The presence of bugs in a software release has become inevitable. The loss incurred by a company due to the presence of bugs in a software release is phenomenal. Modern methods of testing and debugging have shifted focus from “detecting” to “predicting” bugs in the code. The existing models of bug prediction have not been optimized for commercial use. Moreover, the scalability of these models has not been discussed in depth yet. Taking into account the varying costs of fixing bugs, depending on which stage of the software development cycle the bug is detected in, this paper uses two approaches - one model which can be employed when the 'cost of changing code' curve is exponential and the other model can be used otherwise. The cases where each model is best suited are discussed. This paper proposes a model that can be deployed on a cloud platform for software development companies to use. The model in this paper aims to predict the presence or absence of a bug in the code, using machine learning classification models. Using Microsoft Azure's machine learning platform this model can be distributed as a web service worldwide, thus providing Bug Prediction as a Service (BPaaS).