On Efficient Prediction of Indian Stock Indices Using ELM-Based Technique

Abstract

This paper proposes an extreme learning machine (ELM)-based prediction model for forecasting the closing price of two Indian stock indices, Bombay Stock Exchange (BSE) Sensex and National Stock Exchange (NSE) Nifty 50. Using the raw closing price data for the period 2013–2018, ten technical features are extracted from a sliding window of size ten. These features are then used in the ELM model to predict these indices for one, five and fifteen days ahead. To compare the performance of the proposed model the back-propagation artificial neural network (BP-ANN) and trigonometric functional link artificial neural network (TFLANN) models are also simulated. In case of TFLANN and BP-ANN, the convergence characteristics are obtained for the stock indices as well as for different day’s ahead prediction. Similarly, two performance measures, root mean squared error (RMSE) and mean absolute percentage error (MAPE), are obtained for all the three models and for all the three days and for both the indices. Examination of the comparative results indicates that the ELM model is the best prediction model among the three. Hence, it can be concluded that the ELM is the potential candidate for the prediction of stock indices.