Hybrid ARIMA-SVR-MLP Model for Forecasting Nigeria’s Gross Domestic Product (GDP)

Abstract

This paper proposes a novel hybrid ARIMA-SVR-MLP model that integrates Autoregressive Integrated Moving Average (ARIMA), Support Vector Regression (SVR), and Multilayer Perceptron (MLP) neural network to forecast Nigeria's Gross Domestic Product (GDP). The models were fitted on yearly GDP data from 1960-2022. The series was tested for stationarity using the Augmented Dickey-Fuller (ADF) test and found to be stationary at the first differencing. Based on model selection criteria, ARIMA (1, 1, 0) was identified as the appropriate ARIMA model. Support Vector Regression (SVR) was also applied to the original series to capture the nonlinearity that might not have been accounted for by ARIMA. The predictions of the best ARIMA and SVR models were then used as inputs for the MLP neural network to form the hybrid model. Out-of-sample forecasts demonstrate that the hybrid model outperforms individual models in terms of accuracy.