Stock Price Prediction Based on Morphological Similarity Clustering and Hierarchical Temporal Memory

Abstract

Predicting stock prices through historical data is a hot research topic. Stock price data is considered to be a typical time series. Recurrent neural network (RNN), long short-term memory (LSTM), and gated recurrent units (GRU) have been commonly employed to handle this type of data. However, most studies focus on the analysis of individual stocks, thus ignoring the correlation between similar stocks in the entire stock market. This paper proposes a clustering method for mining similar stocks, which combines morphological similarity distance (MSD) and kmeans clustering. Subsequently, Hierarchical Temporal Memory (HTM), an online learning model, is used to learn patterns from similar stocks and make predictions at last, denoted as C-HTM. The experiments on the price prediction show that 1) compared with HTM which has not learned similar stock patterns, C-HTM has better prediction accuracy, 2) in terms of short-term prediction, the performance of C-HTM is better than all baseline models.

Keywords

• Machine learning
• kmeans
• morphological similarity distance
• hierarchical temporal memory
• stock prediction