Refined Software Defect Prediction Using Enhanced JAYA Optimization and Extreme Learning Machine
Debasish Pradhan, Debendra Muduli, Abu Taha Zamani, Syed Irfan Yaqoob, Sultan M. Alanazi, Rakesh Ranjan Kumar, Nikhat Parveen, Mohammad Shameem
Abstract
Ensuring the dependability of software before its public release is of utmost importance. Many software issues arise due to human errors made throughout the development process, highlighting the importance of addressing these errors early. It is crucial to incorporate testing resources at the beginning of development to minimize potential issues. Utilizing an approach that identifies modules susceptible to errors helps potential problems. With an understanding of the significance of precisely anticipating module failures, multiple automated solutions are already emerging. This work presents a refined software defect prediction model that utilizes a meta-heuristic optimization technique. The methodology integrates NASA’s data collection procedure, which involves data cleansing, reducing the dimensionality of features, and predicting defects. Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) are used to decrease the feature dimensionality, while an Extreme Learning Machine (ELM) is utilized for forecasting defects. The ELM parameters, such as weight and biases, are ideally chosen using the suggested improved JAYA optimization (IMJAYA) method. The model’s validation involves assessing its accuracy, sensitivity, specificity, F1 score, and MCC metrics using a <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$10\times 5$ </tex-math></inline-formula> cross-validation. The model is verified using NASA datasets that consist of several classes, such as CM1, KC2, KC3, MC1, PC1, and JM1. The PCA-LDA+ IMJAYA-ELM model yields defect prediction accuracies of 95.73%, 98.08%, 94.87%, 96.23%, 97.10%, and 97.46% for the CM1, KC2, KC3, MC1, PC1, and JM1 datasets, respectively. The research outcomes show encouraging outcomes when using a meta-heuristic optimization technique with smaller feature sets for studies on predicting software defects.