A Novel Approach for Stock Price Prediction Using Gradient Boosting Machine with Feature Engineering (GBM-wFE)

Abstract

The prediction of stock prices has become an exciting area for researchers as well as academicians due to its economic impact and potential business profits. This study proposes a novel multiclass classification ensemble learning approach for predicting stock prices based on historical data using feature engineering. The proposed approach comprises four main steps, which are pre-processing, feature selection, feature engineering, and ensemble methods. We use 11 datasets from Nasdaq and S&P 500 to ensure the accuracy of the proposed approach. Furthermore, eight feature selection algorithms are studied and implemented. More importantly, a feature engineering concept is applied to construct two new features, which are appears to be very auspicious in terms of improving classification accuracy, and this is considered the first study to use feature engineering for multiclass classification using ensemble methods. Finally, seven ensemble machine learning (ML) algorithms are used and compared to discover the ultimate collaboration prediction model. Besides, the best feature selection algorithm is proposed. This study proposes a novel multiclass classification approach called Gradient Boosting Machine with Feature Engineering (GBM-wFE) and Principal Component Analysis (PCA) as the feature selection. We find that GBM-wFE outperforms the previous studies and the overall prediction results are auspicious, as MAPE of 0.0406% is achieved, which is considered the best result compared to the available studies in the literature.

Deep Neural Network and Time Series Approach for Finance Systems

“…Gradient Boosting was adopted by Nabi (2020) to predict the stock market movement with a feature engineering variation (GB-FE) from historical data for the Nasdaq and S&P500. A mean square error (MSE) of 0.041% is attained, proving its effectiveness.…”

ESG Investing: A Sentiment Analysis Approach

“…The work of Vijh et al (2020) has shown that popular machine learning models such as random forests and neural networks are efficient in predicting US stock prices. Nabi et al (2020) and Yang et al (2020) also tried several popular machine-learning approaches to find the superiority of the Gradient Boosting models among others in stock price forecasting.…”

An explainable prediction framework for engineering problems: case studies in reinforced concrete members modeling

“…Emerging as one of the most contemporary machine learning techniques, gradient boosting has shown success in various areas including stock price prediction (Nabi et al, 2020), traffic speed forecast (Zhan et al, 2020), Alzheimer diagnosis (Liu et al, 2020) and health monitoring systems (Tahmassebi et al, 2020). In addition to this, gradient boosting has recently shown promising use in several engineering problems such as automatic detection of cracks from concrete surface (Chun et al, 2021), structural damage assessment for proper maintenance (Chun et al, 2020a, b), prediction of undrained shear strength (Zhang et al, 2021) and safety evaluation of steel trusses (Truong et al, 2020), which opens new avenue in modeling engineering problems including seismic damage assessment of structures.…”