Machine Learning Models for Forecasting of Individual Stocks Price Patterns

Machine Learning Models for Forecasting of Individual Stocks Price Patterns

Dilip Singh Sisodia (National Institute of Technology Raipur, India) and Sagar Jadhav (National Institute of Technology Raipur, India)
DOI: 10.4018/978-1-5225-3870-7.ch008

Abstract

Stock investors always consider potential future prices before investing in any stock for making a profit. A large number of studies are found on the prediction of stock market indices. However, the focus on individual stock closing price predictions well ahead of time is limited. In this chapter, a comparative study of machine-learning-based models is used for the prediction of the closing price of a particular stock. The proposed models are designed using back propagation neural networks (BPNN), support vector regression (SVR) with SMOReg, and linear regression (LR) for the prediction of the closing price of individual stocks. A total of 37 technical indicators (features) derived from historical closing prices of stocks are considered for predicting the future price of stock in a time window of five days. The experiment is performed on stocks listed on Bombay Stock Exchange (BSS), India. The model is trained and tested using feature values extracted from the past five-year closing price of stocks of different sectors including aviation, pharma, banking, entertainment, and IT.
Chapter Preview
Top

Background

Since the advent of the stock market concept, many researchers have applied various machine learning techniques to predict the stock market indices and future stock prices. Some noteworthy contributions are k-nearest neighbor (kNN) (P. C. Chang, Fan, & Liu, 2009), neural networks (NNs) (Schierholt & Dagli, 1996), genetic algorithms (GAs)(K. Kim & Han, 2000) and (Kwon & Moon, 2007), support vector machines (SVMs) (Smola & Schölkopf, 2004), least square SVM (LS-SVM) (Ou & Wang, 2009; L. Yu, Chen, Wang, & Lai, 2009), bacterial chemotaxis Optimization (BCO) (Yudong & Lenan, 2009), rough set-based pseudo outer-product (RSPOP)(Ang & Quek, 2006), and also predict the stock price from news driven models (Gusev et al., n.d.)using sentiment analysis (Trends, Chowdhury, Routh, & Chakrabarti, 2014) and other text mining methods(Mittermayer, 2004). Hybrid techniques like independent component analysis (Lu, 2010) with neural networks, genetic complementary learning (GCL) fuzzy neural network (Tan, Quek, & Ng, 2004) and self-organizing map (SOM) with Fuzzy SVM (f-SVM) (Nguyen & Le, 2014) have also been used for prediction. In (K. Kim & Han, 2000), genetic algorithms (GA) and in (Y. M. Kim et al., 2015) nonparametric model along with artificial neural networks (ANN) have used to predict the Korean stock exchange on a daily basis. In (Afolabi & Olude, 2007) the daily stock prices have predicted by using three different approaches such as back propagation, Kohonen SOM, and a hybrid Kohonen SOM and have proved that hybrid Kohonen is better of the other two.

Key Terms in this Chapter

Stock Indices: The name of stock markets where trading of registered stocks take place such as Bombay Stock Exchange (BSE) and CNX Nifty, etc.

Closing Price: The price of particular stock at the end of trading.

Machine Learning Techniques: A group of methods used to learn the past trading performance of various stocks used in this chapter.

Mean Absolute Error: The average of differences between absolute values of actual stock price and predicted price of stocks.

Time Window: Advance time period used for predicting stock prices.

Opening Price: The price of a stock before start of the trading on a specific day.

Complete Chapter List

Search this Book:
Reset