Effectiveness of Ensemble Machine Learning Algorithms in Weather Forecasting of Bangladesh

Atik Mahabub, Al-Zadid Sultan Bin Habib, M. Rubaiyat Hossain Mondal, Subrato Bharati, Prajoy Podder

Abstract: Machine learning (ML) is considered as a fundamental approach in predicting different phenomena including weather. This paper focuses on the application of ML models in weather forecasting of Bangladesh where the weather changes frequently. The novelty of this work is in the application of ensemble regression algorithms to a raw dataset collected from Bangladesh Meteorological Division for the year 2012 to 2018. The dataset has different attributes, including wind speed, humidity, temperature, and rainfall collected at 33 weather stations across Bangladesh. The dataset is split into training and testing portions; the data for the years 2012 to 2017 is used for training, while the data for the year 2018 is used for testing. The prediction is done using several ML-based regression algorithms including support vector regression (SVR), linear regression, Bayesian ridge, gradient boosting (GB), extreme gradient boosting (XGBoost), category boosting (CatBoost), adaptive boosting (AdaBoost), k-nearest neighbors (KNN) and decision tree regressor (DTR). Our results show that the DTR and CatBoost algorithms outperform the algorithms reported in the literature in terms of mean squared error (MSE), mean absolute percentage error (MAPE), and mean absolute error (MAE).