FORECAST PERFORMANCE OF UNIVARIATE TIME SERIES AND ARTIFICIAL NEURAL NETWORK MODELS

Abstract

In this paper, the better model for forecasting Nigeria monthly Precipitation time series data that exhibit seasonal, periodic variations and non-linearity is determined. The models considered are Seasonal Autoregressive Integrated Moving Average (SARIMA), Fourier Autoregressive (FAR) and Artificial Neural Networks (ANN) models. The accuracy of the out-sample forecast of the model considered was measured based on the following forecast evaluations sum of square error (SSE), Mean square error (MSE) and Root mean square error RMSE. From the results, the FAR model forecast was better than that of SARIMA model based on the values of the forecast evaluations when seasonal and period in the series is considered and ANN model forecast was better that both FAR and SARIMA when the non-linear nature of the precipitation is considered. In conclusion, the FAR model is the most appropriate model for forecasting seasonal and periodic variations while the ANN model is the most suitable model for forecasting non-linearity in Nigeria monthly precipitation time series data