Stacked Ensemble Model for Tropical Cyclone Path Prediction

Abstract

Tropical cyclones (TC) are intense circular storms that cause significant economic and human losses in the coastal areas of the equatorial region. Various statistical models have been proposed to forecast the potential path of TC. This study proposes a stacked ensemble-based method to enhance the effectiveness of predicting TC paths using temporal data. The proposed method can be divided into two phases. In the first phase, the Long Short-Term Memory Networks (LSTM) and Gated Recurrent Unit (GRU) models are optimized with stacked layers to determine the most effective configuration for Stacked LSTM and Stacked GRU. In the second phase, k-fold cross-validation is employed to construct multiple Stacked LSTM and Stacked GRU models, and a Meta learner is used to ensemble the predictions from these models. We evaluate the performance of our proposed model using the temporal China Meteorological Administration (CMA) dataset and compare its results with those obtained from other ensemble and non-ensemble techniques. The results demonstrate a significant reduction in mean square error and variance achieved by the proposed model. The code is available on GitHub: TC path prediction