A novel hybrid JAYA optimization algorithm and extreme learning machine classifier (JAYA-ELM) for software reliability prediction

Abstract

Context:

Software reliability prediction uses historical data analysis to identify software components prone to defects. Effective prediction of software reliability allows for more efficient allocation of testing resources, resulting in more dependable software.

Objective:

It is challenging to identify adequate data features and develop an effective software reliability prediction model due to the complex and uneven class distribution of software defect data. In this research, we propose a computationally efficient and high-performance hybrid machine learning model to address these two issues.

Method:

We present the JAYA-ELM framework, which combines the Jaya optimization algorithm and Extreme Learning Machine (ELM) to predict software defects by selecting optimal features and learning from complex software defect data points. The Jaya optimization algorithm is used in the first phase to extract representative data features from the datasets. With the selected optimal features, we then use the ELM model in the next phase to predict defects.

Results:

We validated the proposed model using various datasets, including NASA and Java datasets sourced from the PROMISE repository. Our results demonstrate that the proposed JAYA-ELM hybrid model outperforms several baseline techniques in terms of prediction accuracy. In this study, we validated the effectiveness of our proposed JAYA-ELM hybrid model using a range of datasets, including those from NASA and Java, available in the PROMISE repository. Our results demonstrate that the JAYA-ELM model significantly outperformed several baseline techniques in terms of prediction accuracy. For instance, when applied to the KC2 dataset from the NASA repository, our model achieved an accuracy of 0.835, surpassing the accuracies of 0.794 to 0.806 attained by other models such as ELM-FA, ELM-GA, ELM-PSO, ELM-BESTF, and ELM-GREDDY. Similarly, in comparisons with methods like GA-SVM, PSO-SVM, SVM, WASVM, and GAWhale-SVM, the JAYA-ELM model consistently delivered superior performance. For example, on the MW1 dataset, our model achieved an accuracy of 0.925, outperforming other models, which ranged from 0.917 to 0.923.

Conclusion:

We propose the JAYA-ELM hybrid model as a software reliability prediction framework that considers optimal feature selection and computational performance. The proposed framework outperforms baseline approaches in most cases, as demonstrated by an empirical analysis of 20 software datasets used in this study.