Comparing various machine learning algorithms for time-series prediction

This study demonstrates application of nonlinear autoregressive neural network (NARNN) model for forecasting of intraday stock of S&P Bombay Stock Exchange (BSE) Sensex-50 index, India. NARNNN is an important category of artificial neural network model generally used for multistep time-series forecasting. The model has been trained using three different training algorithms i.e. Bayesian Regularization (BR), Resilient Backpropagation (RB) and Scaled Conjugate Gradient (SCG). The study adopts different training algorithm for time-series prediction of Intraday stocks for a duration of 10-minutes considering training data of past 60 minutes. The precision of the proposed algorithms has been monitored in terms in terms of mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE) and mean absolute percentage error (MAPE). The results clearly highlight superior performance of BR-algorithm giving much lower values of error responses compared to other training algorithms considered in the study.