Comparative Analysis of Machine Learning Algorithms to predict the Tropical Cyclones

Tropical cyclones (TC) are powerful weather phenomena characterized by low-sea level pressure systems with rotating winds. Accurate classification of these cyclones plays a crucial role in predicting their intensity, potential damage, and issuing timely warnings. In this paper, we explore and compare various machine learning model’s accuracy in the classification of tropical cyclones based on multiple meteorological features, including sea level pressure (SLP) drop, maximum sustained surface winds (MSW), estimated central sea level pressure from centre of the eye of the tropical cyclones, latitude, & longitude are 5 features. The following algorithms are used for comparative analysis of tropical cyclone using best track data: Random Forest, C4.5 Decision Tree, Logistic Regression, Nearest Centroid Classifier, Extreme Learning Machine, Linear Discriminant Analysis, Quadratic Discriminant Analysis, and Long Short-Term Memory. When we use the above said 5 features, it is observed that the Random Forest is obtained the maximum accuracy of 99.10%, followed closely by C4.5 decision tree with an accuracy of 98.54%. Logistic Regression obtained accuracies of 94.06%. LDA, QDA, NCC, LSTM and ELM achieved accuracies of 92.94%, 98.09%, 83.76%, 50.62% and 20.5%, respectively.