A Hybrid Approach of Metaheuristic Algorithms and Group Method of Data Handling to Predict Source Code Testability

Abstract

Testability allows developers to improve software by enabling effective testing throughout the development process. The purpose of this paper is a study based on the prediction of software testability using machine learning (ML) algorithms. Based on experimental data, a group method of data handling (GMDH) algorithm has been proposed to forecast the testability of source code. Metaheuristic algorithms like Ant Colony Optimization (ACO), Firefly Algorithm (FA), and Colliding Bodies Optimization (CBO) have been used to optimize the GMDH parameters in order to enhance and raise the forecasting accuracy. Metaheuristics enhance GMDH, forming diverse hybrids. Comparing eight indicators, CBO achieves top efficiency in software testability prediction with R: 0.674, RMSE: 0.208, MAE: 0.167, and RAE: 1.020. Also, with the results obtained from other indicators, it is determined that it has the best performance. Hence, the CBO algorithm applied to the GMDH model to predict testability is selected as the best method.