AN AI-DRIVEN MACHINE LEARNING FRAMEWORK FOR SUSTAINABLE DEVELOPMENT: ENERGY EFFICIENCY, CLIMATE MODELLING, AND INTELLIGENT WASTE CLASSIFICATIONAN AI-DRIVEN MACHINE LEARNING FRAMEWORK FOR SUSTAINABLE DEVELOPMENT: ENERGY EFFICIENCY, CLIMATE MODELLING, AND I

Abstract

In this research paper, the design, implementation, and evaluation of a all-encompassing machine learning system to support sustainable development is conducted, covering three interrelated areas, including energy efficiency forecasting at the household level, downscaling climate change and detecting extreme events through deep learning, and automated waste classification with convolutional neural networks. It has been applied to real-world data using Python and TensorFlow, Scikit-learn, and PyTorch, such as the UCI Household Power Consumption dataset (2,049,280 samples), ERA5 reanalysis climate data, and the joint TrashNet/TACO waste image dataset (4,000+ labeled images). An forecaster of standardised hourly energy yields RMSE = 0.0237 (R² = 0.9995) with a Random Forest Regressor and a stable sequential learning MSE = 0.385 at 50 training epochs with a network based on an LSTM. A CNN-LSTM hybrid model decreases climate downscaling RMSE by 63.2% compared to bilinear interpolation baselines (R 2 = 0.924). A fine-tuned ResNet-50 is over 90 percent waste classification accuracy (weighted F1 = 0.902) six waste types. Analyses of SHAP and LIME indicate that scientific validity of model outputs are valid, and autoregressive lag features explain more than 42 percent of the importance of energy predictions. The integrated framework specifically responds to SDG 7, SDG 12 and SDG 13 of the UN Sustainable Development Goals, offering a deployable, modular and interpretable AI system to sustainable infrastructure management.