Design of Hybrid Classifiers for Time Series Prediction using Stock Market Data

Abstract

There are a variety of theories, procedures, and approaches that make time series prediction more accurate, demanding, and exciting for researchers. Dynamic and volatile nature of stock data brings it into a sequential data series known as time series data. Stipulated time interval and stochastic behavior in data pattern makes stock market as a best fit use case for time series analysis (TSA). Deep learning (DL) approaches have made tremendous progress in anticipating stock market patterns, and they continue to pique the interest of market traders and investors. The concept of hidden layer makes DL models best fit for any time series data like the stock market. High level validity of accuracy in data brings LSTM as the most preferable approach. LSTM is a complex computational network that works on the concept of RNN. RNN and its variants are also well-known models that handle missing values and noisy data available in a dataset. LSTM is one of the best ANN topologies that deals with function mapping problems and non-linear dynamics. LSTM handles long term dependencies with its gate enabled framework. Nature inspired and evolutionary algorithms help to optimize the selection of parameters and provide stability between performance and complexity of the discussed models. Such algorithms like Artificial Bee Colony (ABC) bring significant improvements in stock forecasting accuracy.