Recurrent dictionary learning for state-space models with an application in stock forecasting

In this work, we introduce a new modeling and inferential tool for dynamical processing of time series. The approach is called recurrent dictionary learning (RDL). proposed model reads as linear Gaussian Markovian state-space involving two operators, the state evolution observation matrices, that assumed to be unknown. These unknown operators (that can seen interpreted dictionaries) sequence hidden states are jointly learnt via an expectation–maximization algorithm. RDL gathers several advantages, namely online processing, probabilistic inference, high expressiveness which usually typical neural networks. particularly well suited stock forecasting. Its performance illustrated on problems: next day forecasting (regression problem) trading (classification problem), given past market observations. Experimental results show our method excels over state-of-the-art analysis models such CNN-TA, MFNN, LSTM.