FF-HMM: A Novel Fluctuation Fusion based Markov Model for Stock Prediction

An accurate stock prediction model will guide a correct ahead action, which will bring huge benefits. To achieve such a goal, researchers have been working on developing an accurate machine learning and deep learning model, considering the promising performance of artificial intelligence models. However, the current methods still have some potential to explore. In this paper, we proposed a novel fluctuation Fusion based Hidden Markov Model stock prediction method (FF-HMM). Instead of only using the original price for model training, our method adopts stochastic fluctuation of the price for training. Besides, the predicted fluctuation is used to adjust the original prediction with the Exponential Moving Average mechanism. The proposed method is compared with 4 typical stock prediction model (i.e, Support Vector Regression, Recurrent Neural Network, Long Short-Term Memory Network, and traditional Hidden Markov Model) on 6 open sourced benchmarks (i.e.,APPLE, AMZN, BA, GOOGLE, UNH, UNQLO). The promising experiment results show that our method performs better than the others, which confirms the feasibility and scalability of our method.