Adaptive Ensemble Learning Model-Based Binary White Shark Optimizer for Software Defect Classification

Software dominates modern enterprises, affecting numerous functions. Software firms constantly experiment with new methodologies to define and assess software quality to stay competitive and ensure excellence. Software engineering uses fundamentals and cutting-edge technology to develop great software. In recent decades, Data-mining techniques and machine learning for classifying problematic software projects have emerged to improve software quality. ML approaches, especially ensemble learning models, are becoming fundamental to software engineers' daily jobs. This work created a binary white shark optimizer (WSO) to optimize standard ensemble learning models. The objective is to identify the most suitable ensemble number for weak learners to maximize accuracy on benchmark datasets. The EM model uses 14 weak learners. Twenty-one experimental runs are performed on 15 software-defective module datasets. The optimized ensemble model outperforms the standard Ensemble learning model in AUC-ROC, Accuracy, Precision, Recall, F1-Score, and Specificity. The enhanced model has an average accuracy of 86%, compared to 76% for the standard ensemble model across all datasets. The optimized model outperformed the conventional ensemble for the same datasets, with an average AUC of 72% compared to 61% for the standard ensemble. The optimized model was more stable than the standard model, with an STD of 5.53E-03 vs 7.24E-02 for the ensemble model. The WSO optimization process strengthens and generalizes optimizeels. The study suggests that evolutionary metaheuristic approaches can enhance EM models' accuracy, trustworthiness, and adaptability.