Prediction Level of Software Maintainability Using Ensemble Method and Feature Selection

Software maintainability is a key external attribute of software quality that assesses how effectively and efficiently software can be modified by those who maintain it. The maintainability level is predicted using a machine learning model based on several software quality attributes, which can then be used to support decision-making during the software maintenance process. The previous research revealed that the generated predictive models for software maintainability levels still do not meet the established accuracy standards. This study discusses machine learning models built using several individual models, such as Lasso Regression, KNN, Regression Tree, M5Rules, SVM, and ANN, along with ensemble methods like Bagging and AdaBoost. Additionally, feature selection techniques are considered to identify the best features to improve the performance of the software maintainability prediction model. This study aims to investigate the performance of machine learning models in various datasets. The performance is evaluated using three metrics: MMRE, MAE, and Pred. The results show that the ANN algorithm is the best in almost all individual models with a score of MMRE 0,88. Ensemble methods have been proven to enhance the performance of models, given that the ensemble methods and individual algorithms used are appropriate. Feature selection techniques can improve some machine learning models by appropriately removing features, and the algorithms used match the data distribution with MMRE 0,84