Reducing errors during stock value prediction Q-Learning based Generic Algorithm

Abstract

Predicting stock values from temporal data requires knowledge about both inter-day and intra-day patterns for the stock. These patterns are usually described with the help of features which includes but are not limited to simple moving average (SMA), exponential moving average (EMA), Bollinger bands, etc. Analyzing changes in these patterns, allows analysts to predict the future trends for these stocks. Based on these predicted future trends, the future value of the stock can be predicted. In this text a genetic algorithm inspired feature selection algorithm is combined with a Q-learning based weighted sum method. This fusion assists in reducing stock prediction errors for both long term and short-term temporal data. It is observed that the proposed method reduces the error of prediction by at least 10% when compared with state-of-the-art methods like neural network prediction and genetic algorithm-based prediction.