EVALUATING THE PERFORMANCE OF MACHINE LEARNING ALGORITHMS FOR PREDICTING METEOROLOGICAL PARAMETERS

Abstract

Artificial learning techniques are currently used for weather and climate forecasting, etc. In this paper, we will evaluate three algorithms for predicting meteorological parameters based on the humidity parameter. Our dataset was taken from Mbujimayi airport in the DRC. So, having the real-world data at our disposal, we used these Machine Learning tools to interpret and understand what happened by training the three models separately, and draw the conclusion as to which was the best model. Then, we used the three models to make some predictions about what our environment will be like tomorrow, and to draw conclusions and make decisions about whether or not our climate is already facing climate change. Three models are used: Decision tree, k-nearest neighbor and neural network, the analysis reveals that of the three models tested, the decision tree scored 81.8% after training with an average prediction of 71.5%, in second place we have the K-nearest neighbor with a score of 70% after training with an average prediction of 70, 8% and the cloture neural network with 64% training and an average prediction of 66.1%. Thus, the decision tree outperforms the other models in terms of training and prediction of meteorological parameters, and is the best model with a very high performance compared to the other models.