TY - JOUR

T1 - Fusion of multiple diverse predictors in stock market

AU - Barak, Sasan

AU - Arjmand, Azadeh

AU - Ortobelli, Sergio

N1 - This is the author’s version of a work that was accepted for publication in Information Fusion. Changes resulting from the publishing process, such as peer review, editing, corrections, structural formatting, and other quality control mechanisms may not be reflected in this document. Changes may have been made to this work since it was submitted for publication. A definitive version was subsequently published in Information Fusion, 36, 2017 DOI: 10.1016/j.inffus.2016.11.006

PY - 2017/7/1

Y1 - 2017/7/1

N2 - Forecasting stock returns and their risk represents one of the most important concerns of market decision makers. Although many studies have examined single classifiers of stock returns and risk methods, fusion methods, which have only recently emerged, require further study in this area. The main aim of this paper is to propose a fusion model based on the use of multiple diverse base classifiers that operate on a common input and a Meta classifier that learns from base classifiers’ outputs to obtain more precise stock return and risk predictions. A set of diversity methods, including Bagging, Boosting and AdaBoost, is applied to create diversity in classifier combinations. Moreover, the number and procedure for selecting base classifiers for fusion schemes is determined using a methodology based on dataset clustering and candidate classifiers’ accuracy. The results demonstrate that Bagging exhibited superior performance within the fusion scheme and could achieve a maximum of 83.6% accuracy with Decision Tree, LAD Tree and Rep Tree for return prediction and 88.2% accuracy with BF Tree, DTNB and LAD Tree in risk prediction. For feature selection part, a wrapper-GA algorithm is developed and compared with the fusion model. This paper seeks to help researcher select the best individual classifiers and fuse the proper scheme in stock market prediction. To illustrate the approach, we apply it to Tehran Stock Exchange (TSE) data for the period from 2002 to 2012.

AB - Forecasting stock returns and their risk represents one of the most important concerns of market decision makers. Although many studies have examined single classifiers of stock returns and risk methods, fusion methods, which have only recently emerged, require further study in this area. The main aim of this paper is to propose a fusion model based on the use of multiple diverse base classifiers that operate on a common input and a Meta classifier that learns from base classifiers’ outputs to obtain more precise stock return and risk predictions. A set of diversity methods, including Bagging, Boosting and AdaBoost, is applied to create diversity in classifier combinations. Moreover, the number and procedure for selecting base classifiers for fusion schemes is determined using a methodology based on dataset clustering and candidate classifiers’ accuracy. The results demonstrate that Bagging exhibited superior performance within the fusion scheme and could achieve a maximum of 83.6% accuracy with Decision Tree, LAD Tree and Rep Tree for return prediction and 88.2% accuracy with BF Tree, DTNB and LAD Tree in risk prediction. For feature selection part, a wrapper-GA algorithm is developed and compared with the fusion model. This paper seeks to help researcher select the best individual classifiers and fuse the proper scheme in stock market prediction. To illustrate the approach, we apply it to Tehran Stock Exchange (TSE) data for the period from 2002 to 2012.

KW - Classifier fusion

KW - Diversity creation

KW - Machine learning

KW - Fundamental analysis

KW - Stock returns prediction

KW - Risk prediction

U2 - 10.1016/j.inffus.2016.11.006

DO - 10.1016/j.inffus.2016.11.006

M3 - Journal article

VL - 36

SP - 90

EP - 102

JO - Information Fusion

JF - Information Fusion

SN - 1566-2535

ER -