Unsupervised Learning Part-Based Representation for Stocks Market Prediction

Machine learning has been extensively studied for its potentials in prediction of financial markets. Popular predicting models, such as logistic regression, are quite effective in predicting the trend of a stock. A robust classifier can maximize the profit of stock purchase while keep the risk low. In this paper, we present a theoretical and empirical framework to apply L2-regularized logistic regression to predict the stock market. Firstly, we select ten factors that may influence the stock trend, and use logistics regression to model the relationship of these factors and the quality of stocks. Instead of directly using logistic regression on the original data, we firstly used NMF to unsupervised learn part-based representations of the data, and then a classifier is trained on those representations. Our experimental results suggest that logistic regression with part-based representation of data is a powerful predictive tool for stock predictions in the financial market.