Leveraging Machine Learning based Ensemble Time Series Prediction Model for Rainfall Using SVM, KNN and Advanced ARIMA+ E-GARCH

Abstract

Today's precipitation is growing increasingly variable, making forecasting increasingly difficult. The Indian Meteorological Department (IMD) currently employs Composite and Stochastic approaches to forecast spring storm precipitation in Asia. As a corollary, planners are unlikely to predict the macroeconomic effects of disasters (due to excessive precipitation) or famine (less precipitation). The amount of precipitation that drops dependent on a variety of factors, including the temperature of the atmosphere, humidity, velocity, mobility, and weather conditions. This paper would then employ the Hybrid time-series predictive ARIMA+ E-GARCH (Exponential Generalized Auto-Regressive Conditional Heteroskedasticity) to predict precise runoff by taking into account different climatic considerations such as maritime tension, water content, relative dampness, min-max heat, heavy ice, geostrophic tallness, breeze patterns, soil dampness, and barometric force. In perspective of RMSE, MAE, and MSE, the proposed hybrid ARIMA+E-GARCH paradigm outperformed single simulations and latest hybrid techniques.