A Novel Stock Index Prediction Approach by Combining Extreme Learning Machine With Symbiosis Organisms Search Algorithm

Abstract

This paper presents a robust, effective metaheuristic technique called Symbiotic organisms search (SOS) algorithm which is incorporated with extreme learning machines (ELM) model to enhance the forecasting performance in the stock market. The stock market is a time series data which is highly uncertain and volatile in nature. Due to ELM's fast learning process and high accuracy, it is mostly used in regression and classification problems in regards to other traditional methods in a neural network. ELM is a supervised learning algorithm that depends upon the number of hidden neurons and also the weights and biases within the input and hidden layer. Selection of appropriate number of hidden neurons decides the prediction ability of ELM model. Therefore, in this article initially, the ELM model is run several times with different numbers of neurons in hidden layer to suitably fix the number of hidden neurons for three different stock indices. Then metaheuristic techniques such as SOS, teaching learning based optimization (TLBO), differential evolution (DE), and particle swarm optimization (PSO) are implemented to optimally design the weights and biases of EML models. The performance of SOS-ELM is compared with that of TLBO-ELM, DE-ELM, and PSO-ELM in predicting the next day closing price of stock indices. Several statistical measures such as MSE (mean square error), MAPE (mean absolute percentage error), and accuracy are used as performance measures. A parametric test called paired sample t-test is used to show the effectiveness of SOS-ELM model over other methods.