COMPARATIVE ANALYSIS OF THE PERFORMANCE OF ARTIFICIAL NEURAL NETWORKS (ANNs) AND AUTOREGRESSIVE INTEGRATED MOVING AVERAGE (ARIMA) MODELS ON RAINFALL FORECASTING

Abstract

The forecasting of the occurrence of events such as a social phenomenon, a natural disaster, a physical observation, personal research, or otherwise based on historical data has helped individuals and organizations in making informed decisions and adequate arrangements for any eventuality that might occur. Rainfall is a physical event that occurs at certain periods, depending upon the geographical location. In the North West of Nigeria, rainfalls usually occur during the months of April to October. At earlier and late stages, the rainfall is usually characterized by strong winds which causes damages to houses, electricity installations and other monumental structures. Draught and flood are other problems associated with rainfall in the North West region of Nigeria. In this paper, two separate tools were employed to forecast the yearly rainfall of Kaduna metropolis which has suffered from severe problems of draught as well as flood in time past. The results obtained showed that Artificial Neural Network (ANN) technique outperformed Autoregressive Integrated Moving Average (ARIMA) technique in cyclical seasonal behavior with a minimum Mean Absolute Percentage Error (MAPE).