Machine Learning Models for Static Energy-Consumption Prediction in Model-View-Controller Applications

This repository contains the dataset, trained machine learning models, preprocessing components, and supporting architectural analysis used in the study: Machine Learning Models for Static Energy-Consumption Prediction in Model–View–Controller Applications. The study investigates static prediction approaches for estimating software energy consumption without executing applications. Unlike traditional measurement-based methods, predictions are derived from software features including size metrics, cyclomatic complexity and duplicated lines. The research focuses on applications developed following the Model–View–Controller (MVC) architectural pattern. A comparative evaluation was conducted between a baseline linear regression model and four non-linear machine learning models: Random Forest, Gradient Boosting, k-Nearest Neighbors and Decision Tree. Results demonstrate that non-linear models significantly improve prediction accuracy compared to the linear baseline, while introducing a trade-off between predictive performance and computational sustainability due to additional training and inference costs. Cyclomatic complexity and architectural coupling were identified as key factors influencing prediction performance. This Zenodo repository includes the following materials: DatasetComplete dataset used for training and evaluation of the predictive models, including software metrics and energy consumption measurements.File: Dataset Apps.xlsx Trained Machine Learning ModelsFour non-linear models trained for direct inference. All models are serialized Python objects (pickle format) and can be loaded in Python for prediction on new data: Random Forest (RF_trained.pkl) Gradient Boosting (GB_trained.pkl) k-Nearest Neighbors (KNN_trained.pkl) Decision Tree (DT_trained.pkl) Feature ScalerMinMaxScaler used during model training to ensure consistent preprocessing of input features. Required for inference on new data.File: minmax_scaler.pkl Applications Architecture AnalysisVisual representation of the architectural analysis conducted on the evaluated MVC applications, illustrating architectural coupling metrics.File: Applications Architecture Analysis.png Decision Tree Rules VisualizationRule-based visualization of the Decision Tree model’s decision logic, enabling interpretability and independent validation without executing the trained model.File: DT_rules.png