Software Defect Prediction Under the Impact of Deep Learning Algorithms

Abstract

Early software flaw detection is crucial for increasing software quality and lowering the costs, time, and effort required for software development. Software faults prediction (SFP) has made extensive use of machine learning (ML), and ML techniques provide a variety of benefits.
results for fault prediction in software. Deep learning excels in a number of fields, including voice recognition, computer vision, and natural language processing. In this research, the performance of two deep learning algorithms—Multi-layer perceptrons (MLPs) and Convolutional Neural Networks (CNN)—is examined in order to address potential influencing variables.
The results of the experiment demonstrate how changing certain factors directly affects the improvement that occurs; these parameters are changed until the right value is found for each one. Additionally, the results demonstrate that the impact of changing the parameters had a significant impact on prediction accuracy, which increased significantly in comparison to the conventional ML algorithm. The tests are run on four widely used NASA datasets to verify our presumptions. The outcome demonstrates how the parameters addressed may boost or lower the measurement of the fault detection rate. Up to 43.5% for PC1, 8% for KC1, 18% for KC2, and 76.5% for CM1 showed improvement.