An optimized feature reduction based currency forecasting model exploring the online sequential extreme learning machine and krill herd strategies

For the prediction of exchange rate, this paper proposes a hybrid learning frame work model which is a joint estimation of On-Line Sequential Extreme Learning Machine (OS-ELM) along with optimized feature reduction using Krill Herd (KH). The proposed learning scheme is compared with Extreme Learning Machine (ELM) and Recurrent Back Propagation Neural Network (RBPNN), considering three factors such as; without feature reduction, with statistical based feature reduction using Principal Component Analysis (PCA) and with optimized feature reduction techniques such as KH, Bacteria Foraging Optimization (BFO) and Particle Swarm Optimization (PSO). The models are applied over USD/INR, USD/EURO, YEN/INR and SGD/INR, constructed using technical indicators and statistical measures considering 3, 5, 7, 12 and 15 as window sizes. The results of comparisons of different performance measures in testing phase and MSE in training process demonstrate that the proposed OSELM-KH exchange rate prediction model is potentiality superior compared to others.