Establishing Ensemble Learning Models for Daily Rainfall Forecasting

The socio-agricultural landscape of a country is greatly influenced by heavy rainfall and precipitation. Forecasting future rainfall is a gift of estimating with many useful applications; it is a major indication of regional topology, climate change, and natural calamities. Prediction and estimation of rainfall are greatly aided by machine learning's remarkable toolbox. This paper's overarching goal is to determine how a subset of machine learning called ensemble learning influences a dataset used for rainfall prediction in order to make the models more accurate. Using an Australian rainfall dataset, the classification techniques presented in this work were first evaluated independently, and subsequently with pragmatic ensemble strategies such as bagging and boosting. In this study, we will examine the rise or fall of the aforementioned measures and show how ensemble learning techniques can reduce bias and variation. Boosting reduced variance by 13.6% on average and bias by 6% overall, according to the study. A generalization of the model's performance was noted, with a false negative rate reduction of over 20% on average. Further optimization of the model's performance can be achieved by applying the methods discussed in this work to the process of hyper-parameter tuning.