A Comparative Study of Tropical Cyclone Prediction using Machine Learning

Tropical cyclones (TC), considered extreme weather events, can induce massive losses in coastal areas. It affects millions of people and cattle and can cause a lot of economic losses. Traditionally numerical weather prediction (NWP), known as Tropical cyclones genesis forecasting, has been conducted based on statistical analysis. Over the last century, many studies have been undertaken on predicting tropical cyclones using machine learning approaches. Nevertheless, none of the papers has given a concrete idea about the best algorithm to predict the TC. In this paper, we have done a comparison among various machine learning algorithms to predict TC, using a numerical dataset. We have taken a number of 12 algorithms, i.e. Gradient Boosting, Adaptive Boosting(AdaBoost), Extreme Gradient Boosting (XGBoost), CatBoost, Decision Tree Regressor (DTR), KNN, Bayesian Ridge Regression, Random Forest, Logistic Regression, Multiple linear regression, Support Vector Machine (SVM), ANN. Our results show that ensemble methods e.g. boosting algorithms (Gradient Boosting, XGBoost, CatBoost, etc.) give better results compared to other machine learning algorithms.