Development of MCS based-ensemble models using CEEMDAN decomposition and machine intelligence

In this paper, stock price data has been predicted using several state-of-the-art methodologies such as stochastic models, machine learning techniqus, and deep learning algorithms. An efficient decomposition method resonating with these Machine Intelligence (MI) models has been embedded with boosting ensemble method. Finally a Model Confidence Set (MCS) based algorithm has been proposed for forecasting stock price data. Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) decomposed orthogonal subseries have been predicted using Random Forests (RFs). Then Kernel Ridge Regression (KRR) model is used to combine those predictions to form a hybrid predictor. In addition, improvement in prediction performance has been observed using kernel functions. Adaptive Boosting (AdaBoost) has been found stimulating prediction accuracy of Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models. CEEMDAN has also increased the performance of AdaBoost. Nevertheless, combination of forecasts observed from various methodologies is a good approach for improving the result. Despite optimizing the weights for the combination of all the models, a heuristic MCS-based snuffing of the least important models prior averaging is conceded as a potent approach. MCS rescinds insignificant models based on the out-of-sample forecasting or in-sample prediction performance prior to equally average the superior models. The proposed methodologies have been compared to the existing standalone techniques using several validation measures. However, CEEMDAN with Support Vector Regression (CCEMDAN_SVR) model has been found to be the best predictor in the current scenario.