Utilizing Machine Learning to Predict Flow State from Gameplay Interaction Data

Flow, a profound psychological state associated with optimal experiences, serves as a pivotal gauge of engagement across various contexts. Traditionally evaluated through methods like surveys, interviews, and sampling surveys for qualitative analysis, these approaches aimed to decipher participants' experiences and perceptions. However, these methods are time-intensive and prone to memory and expressive limitations. In response, contemporary research increasingly leans towards physiological indicators for assessing flow. Utilizing physiological signals to detect flow reduces narrative inaccuracies and minimizes sensitivity to participants' subjective awareness, albeit demanding significant time, human resources, and specialized equipment. To surmount these challenges, this study introduces an innovative methodology for efficiently predicting flow states. utilizing gameplay and interaction data as inputs for machine learning models. By employing a real-time strategy game as the experimental environment, participants' gameplay recordings and interaction records are collected and paired with the Flow Short Scale questionnaire to establish a predictive model for flow state. The results demonstrate the success of this approach, achieving a significant prediction accuracy ( $MAE =0.0623$ ) and highlighting a strong correlation between objective gameplay records and subjective flow experiences. This streamlined methodology offers a promising avenue for quantifying and predicting flow state, contributing to a deeper understanding of engagement dynamics in digital environments.