Financial Time Series Forecasting Using Hybrid Evolutionary Extreme Learning Machine

In this research, we train an extreme learning machine (ELM) with a recently proposed parameter-less evolutionary algorithm called the fully informed search algorithm (FISA) and create a new hybrid model termed FISA + ELM. We then apply the FISA + ELM model to forecast the close prices of two major currency exchange rates. For comparison purposes, we also developed another model known as CRO + ELM, where chemical reaction optimization (CRO) is employed to train ELM in the same way and for the same task. To evaluate the performance of both models, we use two error metrics: mean squared error (MSE) and root mean squared error (RMSE). The experimental analysis proves the proposed model, FISA + ELM, produces lower MSE and RMSE values than the alternate model, CRO + ELM. Overall, we observe that FISA + ELM outperforms CRO + ELM.