Software Fault Prediction Using Machine Learning Models and Comparative Analysis

Abstract

Software Testing is an important phase of Software Development Life cycle. Effective software testing helps in identifying faulty modules, but this process becomes very time consuming, especially for large /complex software. Moreover early identification of error prone modules can be useful in producing better quality software. Hence software fault prediction has gained significant attention of the researchers in the recent years. Mainly two types of techniques are being used for it: statistical methods and Machine learning models, out of which recent trend is more inclined towards machine Learning based techniques. Identification of faulty modules is a binary classification problem and machine learning models such as Decision Tree and its variants, Random Forest and Support Vector Machine, are best suited for the fault prediction. This paper attempts to predict software faults using four different classification models. Further, the performance evaluation of these models is also carried out in this paper using accuracy, Precision, Recall, F1-Score and execution time of over 12 datasets, extracted from one of the most commonly used PROMISE repository. Based on these performance metrics comparison of applied four models is done and our comparative analysis indicates that in most of the cases Support Vector Machine model is able to predict software faults more efficiently than the rest in terms of accuracy as well as execution time