FORECASTING WIND DIRECTION IN ALOR SETAR USING MACHINE LEARNING TIME SERIES MODELS WITH TRIGONOMETRIC TRANSFORMATION

Abstract

Forecasting wind direction is inherently challenging due to its circular nature, where conventional numerical models often encounter discontinuities at the 0°/360° boundary. This study compares two modelling strategies for daily wind direction prediction in Alor Setar, Malaysia, using data from 2013–2017. A transformation-based approach and a direct numerical approach are compared for forecasting wind direction to assess their differences. In the transformation-based method, wind direction values are converted into sine and cosine components to preserve circularity, with predictions later reconstructed using inverse trigonometric functions. The direct approach predicts wind direction values without transformation. Three models, Prophet, Random Forest, and Holt-Winters, are applied under both strategies. Model performance is evaluated using time series plots, wind rose diagrams, and angular error metrics, including Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE). Results indicate that the Random Forest model is the best model for forecasting the wind direction in Alor Setar, and the transformation-based approach produces more accurate and stable predictions, effectively capturing directional continuity, while the direct approach yields higher angular errors and fails to replicate the observed wind direction distribution. To our knowledge, this is one of the first studies in Malaysia to systematically apply transformation-based approaches for wind direction forecasting. The findings highlight the practical importance of improved wind direction prediction for renewable energy optimization, aviation safety, and environmental monitoring.