Software Reliability Prediction Using Deep Learning and Feature Selection Algorithms

Abstract

Software reliability is crucial in preventing user issues, financial losses, and reputational damage to companies. Developing accurate models for estimating reliability is imperative. Deep learning, a branch of artificial intelligence, uses neural networks to understand and analyze data, playing a vital role in predicting errors and improving software quality. In this research, Neural Networks (NN), Recurrent Neural Networks (RNN), and Long Short-Term Memory (LSTM) algorithms, along with statistical methods like Chi-square and Regression Coefficient, and intelligent algorithms such as Particle Swarm Optimization (PSO) and Whale Optimization Algorithm (WOA), were employed for feature selection. The results highlighted the superiority of PSO and WOA over traditional methods, with LSTM outperforming other algorithms. Evaluation metrics, including Accuracy, Precision, Recall, and F1-Score, indicated that WOA with LSTM achieved 100% accuracy across datasets. For DS1, accuracy was 97% for all networks, reaching 100% with WOA. DS2 showed accuracy improvement from 80% to 82% with statistical methods and up to 100% with WOA. DS3 demonstrated 99% accuracy with statistical methods and PSO, reaching 100% with WOA. DS4 maintained 99% accuracy with all methods. DS5 exhibited accuracy ranging from 82% to 84%, reaching 100% with WOA. DS6 had accuracy between 78% and 77%, reaching 100% with WOA. This underscores the effectiveness of deep learning, especially with PSO and WOA, in enhancing software reliability.