A Hybrid EMD-ANN Model for Stock Price Prediction

Financial time series such as foreign exchange rate and stock index, in general, exhibit non-linear and non-stationary behavior. Statistical models and machine learning models, often, fail to predict time series with such behavior. Former models are prone to large statistical errors. While machine learning models such as Support Vector Machines (SVM) and Artificial Neural Network (ANN) suffer from the limitations of overfitting and getting stuck in local minima, etc. In this paper, a hybrid model integrating the advantages of Empirical Mode Decomposition (EMD) and ANN is used to predict the short-term forecasts of Nifty stock index. In first stage, EMD is used to decompose the time series into a set of subseries, namely, intrinsic mode function (IMF) and residue component. In the next stage, ANN is used to predict each IMF independently along with residue component. The results show that the hybrid EMD-ANN model outperformed both SVR and ANN models without decomposition.