Article Preview
Top1. Introduction
An accurate prediction of stock market price is a very tough task due to the dynamic, noisy, complex and non-linear nature of stock data. In addition to this, stock prices are affected by many factors such as firm’s policies, political events, economic conditions, investor’s expectations and commodity price indices (Majhi et al., 2014). Therefore, stock market data are characterized by discontinuities and nonlinearities and forecasting stock price is difficult (Hadavandi et al., 2010). Over the past years, many researchers attempted to develop a model for stock market prediction. Stock market prediction methods reported in the literature can be categorized into statistical and soft computing methods. Statistical or linear methods are based on past data and easy to implement. However, statistical methods failed to capture the non regularity underlying in the stock data and provide poor performance. Examples of the statistical methods are ARIMA, ARCH and GARCH (Niu & Wang, 2014; Chung & Shin, 2018).
To prevent the limitations of conventional models, Computational Intelligence (CI) techniques like soft computing methods have been suggested to predict the stock price (Atsalakis & Valavanis, 2009; Kumar Chander, 2018). Several soft computing techniques including Artificial Neural Network (ANN), Support Vector Machine (SVM) and Adaptive Neuro Fuzzy Interference System (ANFIS) are the popular methods for predicting stock market price. Among many soft computing techniques, ANN is one of the strongest soft computing techniques, which can efficiently find the nonlinear relationship between input and output present in the stock data. Moreover, ANN can approximate any complex and nonlinear function to a high degree of accuracy. ANN models such as Multilayer Perceptron (MLP), Radial Basis Function Neural Network (RBFNN), Probabilistic Neural Network (PNN), Functional Link Artificial Neural Network (FLANN), Wavelet Neural Network (WNN) and Recurrent Neural Network (RNN) are popular and commonly used for stock prediction (Naeini et al., 2010; Oliveria et al., 2011).
Most of the ANN based forecasting models suffer from slow convergence and long training or learning time. Therefore, new algorithms that alleviate such limitations are necessary to make accurate predictions. Recently, many bio inspired algorithms like, Genetic Algorithm (GA) (Chung & Shin, 2018), Differential Evolution (DE) (Minakhi et al., 2014), Modified Cuckoo Search (MCS) (Hegazy et al., 2015), Artificial Bee Colony (ABC) (Mustaffa et al., 2014), Particle Swarm Optimization (PSO) (Hung, 2011), Bacterial Foraging Optimization (BFO) (Majhi et al., 2009) and Biogeography Based Optimization (BBO) (Dehghani & Zangeneh, 2018) have been successfully applied to optimize the parameters of ANN. This paper designed a hybrid model using MLP and CSO for stock price prediction. CSO algorithm is a metaheuristic algorithm and used to find the optimal value for weights and bias of MLP. The prediction efficiency of the proposed model evaluated and demonstrated to be superior performance to the other models.
The layout of the paper is as follows: Section II provides a brief review of related work in this field. Section III describes the way in which CSO-MLP is applied for stock market prediction. Section IV presents the selected stocks and technical indicators used. Section V deals with the experimental results and compared to a benchmark models and other existing models. Section VI highlights the findings and provides suggestions for future scope of the work followed by relevant references.