Enhancing Software Fault Localization Using Deep Learning Techniques

Abstract

The goal of software fault localization, a crucial activity in software development, is to locate the flaws in a program's source code. Program analysis and statistical approaches, which are frequently used in traditional fault localization procedures, may not be able to fully capture complicated fault patterns. An innovative method for locating software faults using deep learning techniques is presented in this abstract. Our method makes use of deep neural network technology to extract complex defect patterns and correlations from software execution data. We build a deep learning model to predict the likelihood that each program statement will be incorrect by modelling the execution traces of the program as input sequences. The model can recognise pertinent aspects and identify subtle fault indications since it learns from a broad data set of labelled faults and non-faulty executions. As a result, our study makes a contribution to the field of software fault localization by outlining a novel strategy that makes use of deep learning. Our method presents a potential strategy to increase the accuracy and effectiveness of fault localization, thereby boosting software quality and development productivity. We do this by utilising the benefits of neural networks in capturing complicated fault patterns.