A Comparative Performance Model of Machine Learning Classifiers on Time Series Prediction for Weather Forecasting

(Conference Publication)

Access paper here

Abstract: Machine learning is a booming technical term in every domain of research. The majority of the technical concepts sounds to accomplish classification task in a real-life scenario. In the literature, the huge number of classification tools, it becomes very necessary to justify the performance of machine learning classifiers. This paper describes four classification techniques that are successfully applied for the prediction of the two most significant features for weather forecasting temperature and relative humidity (RH). A brief introduction of the proposed model with four prediction methodologies—ARMA, MLP, SVM and ELANFIS—follows the discriminate ideas that can create the space for such research. The techniques are then compared on a public data set containing the time series of the two parameters: temperature and relative humidity. As per the data statistics, the parameters are registered on an hourly basis and recorded over a field in an Italian city. An elaborating analysis of the results is performed to provide insights into the satisfactory performance of the models.