Enhanced Stock Index Forecasting: Metaheuristic Insights and Extreme Learning Machines Evaluation

Abstract

Forecasting exact stock prices is challenging due to the dynamic and unpredictable nature of financial markets. In this study, we employed an ELM to forecast the closing prices of four prominent stock indices. What sets our research apart is the use of six distinct metaheuristic algorithms, namely CRO, FWA, SAHA, TLBO, FISA and BMO, to optimize the weights and biases of the ELM network. A noteworthy aspect of our approach is the inclusion of three semi-parametric metaheuristics (BMO, FWA and SAHA) alongside three non-parametric ones. This resulted in the development of six unique forecasting models. We conducted a comprehensive analysis, evaluating the concert of these models based on forecasting errors generated during the validation process. To quantify forecasting accuracy, we employed four error metrics, such as MSE, MAPE, RMSE and ARV. The experimental results unequivocally indicate that the non-parametric metaheuristic-based ELM outperformed its semi-parametric counterparts in terms of forecasting error. This finding underscores the efficacy of parametric-less approaches in stock price forecasting. While training an ELM network, the FISA EA proved to be the most effective of the three parametric-less algorithms. Additionally, we scrutinize the distinctions by contrasting the outcomes of the top-performing model with those of many of the most cutting-edge models found in the literature.