ELM-MFO: A New Nature-Inspired Predictive Model for Financial Contagion Modeling of Indian Currency Market

This paper proposes an enhanced hybridized machine learning approach to forecast future price exchange rates such as US Dollar (USD), Great Britain Pound (GBP) and Australian Dollar (AUD) to INR. This hybridized machine learning approach mainly consists of popularly used Extreme Learning Machine (ELM) and different optimization techniques used to optimize ELM parameters using Particle Swarm Optimization (PSO), Gray Wolf Optimization (GWO) and Moth Flame Optimization (MFO). Out of three optimization techniques, it has been observed that ELM with MFO (ELM-MFO) provides best accuracy in the process of forecasting as compared to rest two optimization techniques. The datasets used for experiments have been collected from public platform having different time delay formats such as one day, seven days, fifteen days and thirty days. Also, several technical indicators as well as statistical measures have been used to augment the original currency pair dataset to get a deeper insight of the datasets. From the experimentation, comparison and validation, it has been proved that the proposed ELM-MFO outperforms all the networks used for the comparison and achieves higher accuracy of 97% and 95% considering the overall and average accuracy, respectively, and additionally, the statistical validation through Kappa statistics shows the strong-level agreement with 73.25% of this ELM-MFO for the augmented currency pair datasets with combination of original attribute, TIs and SMs.