Article Preview
Top1. Introduction
Fast data acquisition technology demands an appropriate analysis and prediction mechanism for its handling (Atsalakis & Valavanis 2009). Due to the advancement of internet technology and the processor's capability, the vast amount of data is generated at a fine interval of time (Pal & Kar, 2019). Therefore, fast absorption and processing techniques are required to handle the data generated at a very rapid rate. The advances of ICT (Information and Communication technology) and computing algorithms, open the horizon of collection and analysis of high-frequency data (data in a regular or irregular interval of time) (Pal & Kar, 2019). In the recent years, the developments of machine learning algorithms for data analytics (Mahdavinejad at el. 2018, Kumar 2017, 2018 (a, b)) play an essential role in providing an excellent and fast prediction over a vast amount of big high-frequency data (Calcagnile at el. (2018). The three big data attributes, i.e., three Vs. (velocity, volume, variety) are exhibited by the stock market stochastic dataset. Therefore, accurate forecasting or prediction of stock prices is the primary concern for the investors and companies operating in the stock market. Due to the non-stationary and non-linear time-series nature of stock market trends, the prediction of stock prices is a hugely challenging task in the financial market (Zhang et al. 2018). Economic time series analysis is a significant source of information for stock market prediction. Finding hidden patterns is the requirement of analysis and forecasting for the price actuations (Zhou et al. 2018). Existing frameworks for analysis and forecasting of high-frequency financial data sets can be classified into two categories (Zhou et al. 2018):
- 1.
Statistical models in which advanced mathematical models and procedures can analyze the high-frequency dataset. As the analysis of financial data sets requires some underlying assumptions to be followed, therefore this category of methods can't be utilized to develop an intelligent system.
- 2.
The use of soft computing models based on machine learning approaches to capture the dynamics of a financial dataset in the analysis, like in the stock price prediction (Mahdavinejad at el. 2018, Zhou et al. 2018).
In recent years, many stock market forecasting techniques have been proposed to predict the stochastic stock market data, but the accurate prediction of is still not a fully solved problem (Dai et al.,2013). Due to the slow and complexity in the processing of traditional and fundamental statistical methods, the prediction using analytical tools has a minimal application or obsolete in the analysis of high-frequency stochastic stock data.
The soft computing model has shown the better capability to handle the complex, Brownian, and nonlinear dataset of the stock market (Göçken et al., 2019). The proposed work is focusing on devising and applying the soft computing model or machine learning models in the new scenario. The proposed work selects the three best available models from three different paradigms. Auto-Regressive Integrated Moving Average (ARIMA) (Challa et al., 2018) for best statistical learning, Deep Learning for nonparametric machine learning model (Ding et al., 2015), June) and Gradient Boosting Machine (GBM) to ensemble tree-based machine learning model (Basak et al., 2019). The ARIMA is designed and developed by the ‘Forecast' package in R-studio (Gandrud (2016). The Deep Learning and GBM are designed and developed by H2O package in R-studio that is capable and renowned to handle the big stochastic data.
ARIMA is widely regarded and efficient model used in the analysis and prediction of the stochastic stock market (Rathnayaka et al. 2015). It is a time series model which performs based on the past value of the datasets as well as previous error terms for the forecasting.