Machine Learning for Sustainable Forecasting: Adaptive Wind Speed Prediction Using Functional Data

The global switch to sustainable and clean electricity sources depends heavily on wind energy. To ensure grid stability, minimise operating costs, and optimise the efficiency of wind energy systems, accurate wind speed forecasts is crucial. Using functional data from past weather patterns, this study proposes an adaptive machine learning-based method for wind speed prediction. Time-based indicators, temperature, humidity, atmospheric pressure, dew point, and other important meteorological characteristics are included in the dataset, which was gathered via the Open-Meteo weather API for the years 2024–2025. Advanced preprocessing methods, including feature scaling, correlation analysis, and outlier treatment, along with thorough exploratory data analysis, greatly enhanced the quality of the data and the performance of the model. Standard performance metrics including MAE, MSE, RMSE, and R2 score were used to train and assess a variety of regression models, such as Linear Regression, Random Forest, XGBoost, and LightGBM. When it came to capturing the non-linear patterns of wind speed, ensemble-based models performed better. The results highlight the potential of machine learning models in creating reliable, real-time forecasting systems for sustainable energy planning and validate their efficacy within a functional data horizon. Keywords: Wind Speed Forecasting, Sustainable Forecasting, Ensemble Models XGBoost, Weather Prediction, Open-Meteo API, Regression Model.