Abstract
- Adaptive difficulty adjustment represents a critical mechanism for optimizing user experience in digital games. This study introduces a data-driven approach that integrates game telemetry, biometric sensor data, and personality profiles to classify player behavior, which can be used to inform adaptive artificial intelligence (AI) systems. A custom-built game prototype enables the collection of multidimensional data, including reaction times, accuracy, physiological responses (e.g. heart rate), and subjective stress assessments.Unsupervised machine learning techniques, specifically K-Means and Agglomerative Clustering, were applied in multiple analytical iterations, first to segment players by performance and then by strategic behavior. The dataset was engineered to reflect gameplay efficiency and decision-making patterns, with preprocessing steps to minimize bias. Cluster validity was assessed using ANOVA, and feature relevance was evaluated through Random Forest analysis.Results indicate that distinct clusters can be identified, each characterized by specific cognitive and strategic tendencies. These clusters display unique patterns across gameplay performance, physiological feedback, and subjective perception. The proposed classification model enables the foundation for personalized AI systems, capable of dynamic adaptation based not only on skill level but also on emotional and behavioral traits. The approach contributes to the intersection of game analytics, psychophysiological modeling, and adaptive gameplay design.