ARIMA and RNN for Selection Sequences Prediction in Iowa Gambling Task

Abstract

The Iowa Gambling Task (IGT) has become the classical experiment with many studies of cognitive decision models. In this work, we explore whether Autoregressive Integrated Moving Average (ARIMA) models and Recurrent Neural Networks (RNN) in time series analysis can be applied to extract the decision features of IGT participants. The simulation results of IGT show that both models can capture the selection characteristics of participants and make subsequent selection prediction accordingly. Furthermore, the RNN containing selection features with different preferences can represent the corresponding participants to participate in the IGT experiment.