Authors: Nimisha Gupta; Mitul Kumar Ahirwal; Mithilesh Atulkar
Addresses: Department of Computer Application, NIT Raipur, 492010, CG, India ' Department of Computer Science and Engineering, MANIT, Bhopal, 462003, MP, India ' Department of Computer Application, NIT Raipur, 492010, CG, India
Abstract: In this paper, modelling of human decision making process and comparison among various reinforcement learning (RL) techniques with utility functions has been performed. Iowa gambling task (IGT) is used to collect real time data to understand and model the decision making (DM) process involving uncertainty, risk or ambiguity. Performance of models is evaluated based on their mean square deviation (MSD) value. This helps to predict the probability of the next choice that lead to the selection of the advantageous deck as compared to disadvantageous one. Along with that, the deck selection pattern between male and female with the learning process of the participants were also analysed. By comparing the MSD value of various RL models, it is found that the MSD value of DM model consists of prospect utility (PU)-decay reinforcement learning (DRI) with trial dependent choice (TDC) rule is best.
Keywords: human decision making; Iowa gambling task; IGT; reinforcement learning model; utility functions.
International Journal of Information and Decision Sciences, 2022 Vol.14 No.1, pp.15 - 38
Received: 23 Apr 2020
Accepted: 01 Sep 2020
Published online: 09 May 2022 *