Stacked Ensemble Model for Tropical Cyclone Path Prediction

Abstract

Tropical cyclones are intense circular storms that cause significant economic and life losses in the coastal areas of the equatorial region. Various statistical models were proposed to forecast the tropical cyclone’s potential path. This study has proposed a stacked ensemble-based method to increase temporal data’s Tropical Cyclone path prediction effectiveness. The proposed method can be divided into two phases; in the first phase, Long Short-Term Memory Networks(LSTM) and Gated Recurrent Unit(GRU) are optimized with stacked layers and investigated the best possible stacked layers for Stacked LSTM and Stacked GRU. In the second phase, k-fold cross-validation is used to construct multiple Stacked LSTM and stacked GRU models, and Meta learner is used to ensemble the predictions of the numerous trained models. We investigate our proposed model on the temporal China Meteorological Administration (CMA) dataset and compare the results with other ensemble and non-ensemble-based techniques. Results show an apparent reduction in the proposed model’s mean square error and variance. The code is available on GitHub: TC path prediction.