Short Term Weather Forecasting Comparison Based on Machine Learning Algorithms

Abstract

Forecasting is the term used to describe the attempt to predict outcomes in unknown or uncertain situations. The most vital factor in many applications of weather forecasting is air temperature. The air temperature alone can’t be the effecting point of forecasting weather. Moreover, with the advancement of computer technologies, forecasting models have been transformed widely. This paper approached a system that forecasts air temperature using machine learning algorithms. Several regression methods were employed to attempt to predict temperatures. This research evaluated four algorithms (Decision Tree, AdaBoost, Random Forest, and Gradient Boosting) on some meteorological data over three years (2015-2019), where 80 percent of the total data set was utilized for training and tested on 20 percent. The variables used include Wind speed, Relative Humidity, Dew point, and Air pressure. The objective was to determine which regressor achieves better outcomes for forecasting air temperature with the lowest error rate. This research concluded that the Random Forest Regressor is the most accurate in prediction. Here, MAE is used to determine the accuracy. On average, the Random forest had the lowest MAE value of 0.102, which was lower than the outcomes of the other three algorithms.