Development of a Machine Learning Method for Analyzing Cognitive Bias by Designing a Loss Function, etc. Considering Objective Probability

Abstract

In behavioral economics, when human cognition intervenes in probabilistic events, it is known that there is a cognitive bias that results in a biased perception of probability depending on the content of the cognition. In order to analyze the cause of this bias, a common method is to compare objective probability, in which cognitive bias does not intervene, with subjective probability, in which cognitive bias does intervene. However, there are few events for which objective probability can be estimated, and in previous research, analysis has been conducted by treating subjective probability, which has little intervening bias, as objective probability based on statistical calculations. In addition, it has recently been shown that it is possible to create a learner that is closer to objective probability by selecting input features, in contrast to the conventional statistical approach. However, there is bias in the selection of input features. In this study, we show that it is possible to create a learner with a loss function that indicates the degree to which the learner is able to calculate the objective probability. By using this method, we will discuss the possibility of estimating objective probability that is less susceptible to cognitive bias, as well as the possibility of conducting quantitative analysis.