A Novel Technique for Optimization of Artificial Neural Network Using Ensemble of Chimp, Harris Hawks and Manta Ray Foraging Optimization Algorithms for Enhancing Software Maintainability Prediction

For improving software maintainability prediction (SMP), an efficient method for SMP is required in early stages of SDLC to avoid higher maintenance costs involved with the software. To address this issue, we have used three metaheuristic algorithms, i.e., chimp optimization algorithm (ChOA), Harris hawks optimization (HHO) and manta ray foraging optimization (MRFO) algorithm. Chimp optimization algorithm has two variants: ChOA1 and ChOA2 depending on the dynamic coefficients of h vector. These metaheuristic algorithms are used for artificial neural network (multilayer perceptron) hyperparameters optimization which can be useful for improving the predictive capability of ANN which will then correctly predict software maintainability (change) using various OO metrics. In this study, a novel HCHHMRFO algorithm has been developed to select the best hyperparameters of ANN (multilayer perceptron) to build the model, which can predict software maintainability accurately. Performance evaluation of models developed after hyperparameters tuning with ChOA1, ChOA2, HHO, MRFO, Ensemble of ChOA, HHO and MRFO (ChOA1_HHO_MRFO and ChOA2_HHO_MRFO) is performed using performance indicators including MAE and RMSE. UIMS and QUES datasets have been used to assess this work. Tent map and singer map are the best chaotic maps (in terms of iterations; RMSE; and MAE, respectively) that are used for ANN’s hyperparameters initialization on UIMS dataset. Gauss/mouse map and tent map are the best chaotic maps (in terms of iterations; RMSE and MAE, respectively) that are used for ANN’s hyperparameters initialization on the QUES dataset. Comparison results suggest that ChOA2_HHO_MRFO Ensemble is found to give better results to tune ANN hyperparameters for SMP for both the datasets, i.e., UIMS and QUES.