ChallengeDetect: Investigating the Potential of Detecting In-Game Challenge Experience from Physiological Measures

Title of the Paper

ChallengeDetect: Investigating the Potential of Detecting In-Game Challenge Experience from Physiological Measures

Paper Information

• Subject Area: Game User Experience and Human-Computer Interaction
• Keywords: Perceived challenge, video games, physiological signals, machine learning, player experience

Research Background and Problem Statement

• Identified Issues or Challenges: Many current methods for measuring game experience rely on discrete questionnaire tools (e.g., CORGIS), which cannot provide continuous real-time assessments. Additionally, the integration of physiological signals with game difficulty adjustment has primarily focused on traditional physiological or cognitive challenges, lacking research on the dynamic detection of emotional challenges.
• Significance: Challenge, as a core element in video games, encompasses cognitive, emotional, and physical experiences. Real-time, dynamic detection of challenge experiences can help understand how players' experiences evolve during gameplay, thereby optimizing game design to better adapt to players' skills, experiences, and motivations.
• Research Motivation and Related Work:
  • While the CORGIS questionnaire can distinguish between different types of challenges (cognitive, performance, emotional, and decision-making challenges), it can only assess overall experiences offline.
  • Previous studies have primarily used physiological signals to detect emotions and game difficulty but have not delved deeply into the complex experiences associated with emotional challenges.
  • A review of existing machine learning-based game experience research reveals that various algorithms can identify different player states, but comprehensive detection of different challenge types remains unexplored.

Proposed Solution

• Proposed Methods or Solutions:
  • Design experiments to collect physiological signals and perceived challenge data.
  • Use various machine learning models to detect perceived challenges.
  • Extract key physiological features and analyze their relationships with different types of challenges.
• Innovations:
  • Proposed a comprehensive data processing workflow for detecting multiple types of perceived challenges using physiological signals.
  • Conducted the first dynamic detection of cognitive, performance, emotional, and decision-making challenges, exploring their independence and coexistence.
  • Combined machine learning and feature importance techniques to optimize the accuracy of challenge detection.
• Implementation Steps and Key Techniques:
  • Experimental Design: Used the game Fallout 4 to construct three different game scenarios covering cognitive, performance, and emotional challenge types, while capturing players' experiences in real-time through questionnaires.
  • Data Processing: Extracted features from physiological signals such as electrocardiogram (ECG), electrodermal activity (EDA), respiratory activity (RSP), surface electromyography (EMG), and skin temperature (TEM), and performed preprocessing (e.g., sliding window techniques).
  • Model Training and Evaluation: Applied various machine learning algorithms (e.g., linear regression, support vector machines, and deep learning) for challenge detection.
  • Feature Selection: Analyzed key physiological features using feature importance techniques from linear regression and random forest models.

Research Findings

• Specific Results:
  1. The experimental dataset includes physiological signals and continuous challenge experiences from 32 players across 3 game scenarios.
  2. The detection models achieved an accuracy of approximately 80% for challenge activation in most cases, with a maximum accuracy of 85%.
  3. Identified 24 challenge-related physiological features, revealing significant physiological signals under different types of challenges.
• Advantages:
  • Provides a new approach for dynamically assessing perceived challenges, addressing the limitation of offline questionnaires that cannot continuously monitor experiences.
  • Optimized the application of machine learning methods, achieving high prediction accuracy.
  • The challenge detection method offers a basis for dynamic difficulty adjustment and real-time improvement of game experiences.
• Experimental or Evaluation Results:
  • In 10-fold cross-validation, the deep learning model (DNN4) performed the best, with an average error rate (RMSE) of 0.84.
  • In "leave-one-player-out" cross-validation, simpler models like linear regression and random forest demonstrated higher robustness.
  • Feature importance analysis indicated that electrocardiogram (ECG) and surface electromyography (EMG) signals are the most critical physiological indicators.
• Limitations and Future Directions:
  • The current model does not fully account for the dynamic nature of challenge experiences over time, requiring further research into their temporal characteristics.
  • The dataset is limited to a single game type; future studies should expand to other games and a broader user base.
  • Physiological signals may have certain biases (e.g., high arousal states affecting detection results), and integrating other player behavior data could optimize evaluations.

This study provides a new technical framework for real-time detection of players' perceived challenge experiences, while also highlighting the vast potential in the field of dynamic evaluation of game experiences.