Software maintainability prediction based on change metric using neural network models

The increased software system complexity requires better approaches to advise developers in the process of changing parts of the software during the maintenance and evolution process. The aim of this paper is twofold: (1) to provide an in-depth analysis and comparison of various existing and newly developed maintainability prediction models, and (2) to investigate and obtain the best set of metrics for software maintainability prediction. Various code metrics were employed to explore and discover the best model, three previously developed configuration metrics as dependent variables, i.e. with 10 metrics, 2 metrics, and 3 metrics. The new proposed model considers the correlation between code metrics and the maintenance change metric (four and five metrics were considered in the two used projects). The overall framework of our Neural Network Change Prediction (NNCP) approach uses an artificial neural network algorithm with various input layers. Different validation methods were employed: k-fold cross-validation and hold-out validation. Statistical analysis performed using the Wilcoxon signed rank test between the four models shows that the new proposed model can better predict the maintainability change metric, having one metric in common with the compared model but considering other three distinct ones. • A comparison of developed maintainability prediction models is provided. • The new model considers the correlation between code metrics and the change metric. • An artificial neural network algorithm is employed with different validation methods. • Wilcoxon signed-rank test shows that the model best predicts maintainability change.