Performance Analysis of Ensemble Techniques for Rainfall Prediction: A Study Based on the Current Atmospheric Parameters

Abstract

Rainfall prediction is the most significant requirement nowadays due to the chaotic nature of climate. Climate has changed drastically over the last few years due to global warming and has become very unpredictable. Rainfall prediction is essential for decision-making in various sectors like agriculture, transportation, tours and travels, outdoor events, etc. In this study, machine learning algorithms are analysed and experimented on the dataset comprising various atmospheric parameters of Hyderabad city in Telangana. The work is carried out on individual popular classifiers, namely Naïve Byes, Decision Tree, Random Forest, K nearest neighbour, and Support Vector Machine. The performance is compared with techniques like voting classifiers and stacking ensemble. The experiment gives predictions on the rainfall intensity as No Rain, Low to Medium rain, or Heavy rain. The k-cross-fold validation technique is used as the evaluation metric, which is very effective and results in less biased estimations. The aim is to provide the decision-making capabilities based on the mentioned intensity of rainfall that can be very useful in managing the irrigation cycle in the agriculture field, deciding on an outdoor event, or any travelling plan based on the current atmospheric parameters. The platform used is python, which is portable, open to access, and available easily.