A Comparative Analysis of Univariate Time Series Prediction by Mathematical Models

Abstract

In this paper a comparative study is presented about time series prediction methods using three algorithms: autoregressive (AR), autoregressive integrated moving average (ARIMA) and neural network of type multilayer perceptron trained by backpropagation algorithm. The theoretical basis of time series prediction by building and using of common mathematical models is presented. As the different time series characteristics need different prediction approaches investigation is performed about the details about the predicted values. Experimental results of real time series prediction are summarized and analyzed for prediction of electrical consumption observations. Prototype software modules for the prediction are developed by the author using the programming language Java as software libraries and they can also be integrated and practically used in other software systems.