Application of Neural Network with Extreme Learning Machine

In light of recent economic uncertainty, the accurate forecasting of economic growth has become increasingly important. This requires the consideration of multiple macroeconomic indicators, which may have different frequencies (e.g. monthly, quarterly, yearly). To address this issue, this paper proposes a modified single-hidden layer feedforward neural network (SLFN) that incorporates the mixed data sampling (MIDAS) approach. Furthermore, the high computation cost of the SLFN-MIDAS model is addressed by incorporating the extreme learning machine (ELM) algorithm. The proposed model is then compared to the SLFN with back propagation (BP) learning approach and the classical regression-MIDAS using root mean square error (RMSE) and mean square error (MSE) as evaluation metrics. The results of the experiments indicate that the ELM-based model outperforms the BP-based model and the traditional MIDAS in terms of prediction accuracy for all types of frequency alignment. However, the ELM learning algorithm is found to be slower than the regression-MIDAS algorithm, although it runs around 100 times faster than BP for this empirical study.