Forecasting stock price by hybrid model of cascading Multivariate Adaptive Regression Splines and Deep Neural Network. (October 2021)

• Record Type: Journal Article
• Authors: Bose, Ankita; Hsu, Ching-Hsien; Roy, Sanjiban Sekhar; Lee, Kun Chang; Mohammadi-ivatloo, Behnam; Abimannan, Satheesh
• Is Part Of: Computers & electrical engineering. Volume 95(2021)
• Publication Date: 2021-10
• Subjects: Stock closing price prediction -- Hybrid model -- Multivariate adaptive regression splines -- Deep neural networks -- Correlation; Computer engineering -- Periodicals; Electrical engineering -- Periodicals; Electrical engineering -- Data processing -- Periodicals; Ordinateurs -- Conception et construction -- Périodiques; Électrotechnique -- Périodiques; Électrotechnique -- Informatique -- Périodiques; Computer engineering; Electrical engineering; Electrical engineering -- Data processing; Periodicals; Electronic journals; 621.302854
• DOI: 10.1016/j.compeleceng.2021.107405
• Languages: English
• ISSNs: 0045-7906

Abstract

Highlights: Accurate Stock price prediction can help mitigate the friction that exists in stock investments. Multiple factors contribute to the closing price of stocks so it is important to know which ones are more influential. Applying Multivariate Adaptive Regression Splines (MARS) to remove non-influential parameters proves quite effective. With the immense potential to learn from data, Deep Learning (DL) emerges as a strong regression algorithm to predict the Closing Prices of Stocks. A Hybrid Machine Learning Algorithm that first applies MARS for dimensionality reduction, followed by DL proves to be a highly accurate model. This Hybrid model is low on computation as well. Abstract: Much of the hesitation in stock investments is due to apparent volatility about the stock price. Had there been a predictor to accurately predict the final trading price of stocks, it could be an assurance to invest in the Stock Market. Thus we propose, a trustworthy hybrid model by cascading Multivariate Adaptive Regression Splines(MARS) and Deep Neural Network(DNN), to predict closing prices of stock. The high-frequency KOSPI data set has been used and a customized pre-processing algorithm has been applied to clean the data. MARS is then been applied on this clean data and the attributes retained by MARS are passed to a DNN for training. Such application has resulted up to 92% closing price prediction accuracy. Thus, our hybrid model successfully has reduced the dimensional feature without compromising on accuracy as it gave better results than MARS and DNNs individually. Data-Augmentation has also been used to further verify the outcome of this application. Main metrics used for performance evaluation are Correlation(RHO) and R2 value.