Causality tests using basic and optimally-pruned extreme learning machines

Abstract

This study explores the application of artificial intelligence on the causal relationship between mining production index and electricity load. The data used is the total mining production index and total electricity consumption in the mining sector sampled on a monthly basis from January 1985 to December 2011 in South Africa. Optimally-pruned and basic extreme learning machines were used to develop nonlinear models on the datasets under consideration. This work introduces the use of extreme learning machines for modeling. It was found that, using OP-ELM, there is statistically significant improvement in forecasting error of the electricity load by using both the mining production index and the electricity load on the conditioning set and the same set showed no improvement in predicting the production index. ELM results showed no improvement for both electricity load and production index. It was concluded that the OP-ELM models showed a unidirectional granger causal relationship running from the mining production to electricity load.