A Multi-class Classification Approach for Weather Forecasting with Machine Learning Techniques

Abstract

Weather forecasting is vital as extreme weather events can cause damage and even death. The science of meteorology in recent decades has made spectacular progress resulting in more reliable forecasts. Although meteorologists now have adopted modern tools for accurate weather forecasting, extreme and sudden climate changes in the atmosphere have posed accurate weather forecasting even more valuable. In this research paper, we present a multi-class classification methodology from machine learning (ML) in order to predict the five classes of weather conditions. Specifically, the One-Against-One (OAO) and One-Against-All (OAA) strategies are evaluated under Support Vector Machine (SVM) and Logistic Regression (LR) assuming, for comparison, Random Forest (RF) and k-Nearest Neighbours (k-NN). The prevailing model is linear SVM under the OAO method achieving the average Accuracy, Precision, Recall, F-Measure and Area Under Curve (AUC) of 96.64%, 96.8%, 96.6%, 96.6% and 98.5%, respectively.