Rainfall forecasting using the group method of data handling model: A case study of Sarawak, Malaysia

Abstract

Time series forecasting has led to the emergence of various forecasting models applied to arrays of time series problems, such as rainfall forecasting, dengue forecasting, tourism forecasting, and others. The Artificial Neural Network (ANN) is a popular Artificial Intelligence (AI) model extensively employed in much research for time series forecasting due to its nonlinear modeling ability. The group method of data handling (GMDH) is an AI model with the characteristics of heuristic self-organizing capability. This model has shown successful results in many areas. Nowadays, rainfall forecasting remains a vital interest and is still actively researched, where researchers use different soft computing techniques. The ANN has been popularly studied for rainfall forecasting because of its ability to efficiently train a large amount of data and completely detect complex connections between nonlinear dependent and independent variables. However, research on rainfall forecasting using the GMDH model is limited. Hence, this paper designates the GMDH model and its application to rainfall forecasting. The conventional GMDH model uses the polynomial transfer function. The sigmoid transfer function is proven to solve the multicollinearity issue caused by the quadratic polynomial of the GMDH model. Hence, this research tackled the multicollinearity issue of using different transfer functions in GMDH modeling and forecasting. The study compares the results of using polynomial and sigmoid transfer functions for the GMDH model development. This research uses the Malaysia rainfall dataset of the Sarawak regions from 2010 until 2019 as a case study to evaluate the effectiveness of the GMDH models in this research. The results exhibit that the polynomial transfer function is dominant in achieving the smallest RMSE and MSE values in all regions.