Long Short-Term Memory with Cellular Automata (LSTMCA) for Stock Value Prediction

Abstract

Time series analysis is a difficult task as the data changes dynamically with time. Stock value prediction is one of the examples of highly variable data which cannot be averaged for computation purpose. Even though many approaches are available for predicting stock values, still there is plenty of room for thinking of a better classifier with more accuracy and adaptability. Many statistical analysis methods are available for predicting the stock where the predictability low when tested with different data sets. We have studied the existing literature of both statistic and dynamic classifiers and arrived at a classifier developed with a deep learning technique augmented with cellular automata. We propose a long short-term memory cellular automata (LSTMCA) classifier for news data set in order to predict stock level either increase or decrease in next ten days. We have tested our classifier LSTMCA with existing literature and found an improvement of 2.67% when tested with news data sets.