Using Machine Learning Algorithms in Studies of Human Decision Making under the Continuum and Binary Types of Alternatives

Date

2023-08

Journal Title

Journal ISSN

Volume Title

Publisher

Abstract

The ability of living organisms to make random decisions under conditions of uncertainty may be important for survival. However, decisions made at random often turn out to be ineffective and irrational. The factors that predetermine what we choose could occur either because of our past life experiences or because of our representation as a group in society. In this dissertation, we address the use of statistical and mathematical learning methods to evaluate the patterns of choices made by a group of subjects under a continuum and binary types of alternatives in an unsupervised and supervised learning setting. The study consists of two research projects. The first project aimed to identify common patterns among the choices made by subjects in a psychological experiment. The experiment involved subjects drawing a straight line from the center of a page in any direction on each of 10 pages of a notebook, with the freedom to choose angles randomly without having to repeat what they chose on the previous page. The study analyzed the data using mathe- matical and statistical learning methods, capturing the patterns of choices made by the subjects under a continuum of equivalent alternatives. The results demonstrated that when a group of subjects is faced with subjectively equivalent alternatives, their choices may not be entirely random, but rather influenced by non-essential or irrel- evant features in the environment or by their previous life experiences. Different subjects may focus on different features of the choice situation, leading to variations in the patterns of choices made. The second project aimed to build a prediction model to predict the outcome of the 2022 US House election. The study suggested that representation as a group or a subdivision in society could make a difference in an election decision of the electorate and that a prediction model of the US House election should consider each district as a separate observation and identify the pre- dictor variables that influence the election outcome in the district level. The study used statistical machine learning models to predict the US House election outcome using district-level demographic and socio-economic variables, considering each dis- trict as a separate observation. The predictor variables were selected by considering the most common racial and economic subdivisions in society so that the influence of representation as a group in decision-making can be captured. The study sheds light on the potential reasons behind the patterns observed in both experiments and emphasizes the importance of using statistical and machine learning methods to evaluate the patterns of choices made by individuals and groups. Further research is needed to investigate the generalizability of the findings to differ- ent decision-making scenarios and populations. Nevertheless, the study contributes to the ongoing discourse in experimental psychology by exploring the complex na- ture of decision-making and the influence of various factors on the choices made by individuals and groups.

Description

Keywords

choice overload, Gaussian mixture modeling, behavior patterning, Random forest classification, Election prediction, stochastic neighbor embedding

Citation