Wind Data Pattern and Trend Analysis Using Feature Identification and Extreme Wind Speed Prediction

Abstract

The most dangerous and destructive natural disasters in the world are wind-related. A literature review on machine learning-based approach is done for identification of wind disaster types and the forecasting of extreme wind speed. The study utilizes statistical techniques and machine learning models to uncover valuable insights into wind behavior and develop accurate predictions. A comprehensive dataset of wind speed and direction measurements is collected and preprocessed, ensuring data quality. Relevant features, including meteorological variables, geographical factors, and seasonal indicators, are extracted for feature identification. Predictive models are employed to predict extreme wind speeds resulting in RNN (Accuracy - 0.976), LSTM (Accuracy - 0.979), MLP Classifier (Accuracy - 0.801). The model is verified and the models' performance by comparing predicted extreme wind speeds with observed data, employing metrics like root mean square error (RMSE) or mean absolute error (MAE).