Automated affect classification and task difficulty adaptation in a competitive scenario based on physiological...

Area of Science:

• Human-Computer Interaction
• Affective Computing
• Physiological Computing

Background:

• Dynamic task difficulty adaptation is crucial for competitive and cooperative scenarios.
• Current methods rely on performance, neglecting user-specific factors like workload and affect.
• Physiological linkage offers a novel approach to infer user states for adaptation.

Purpose of the Study:

• To investigate automated affect recognition using physiological linkage for dynamic task difficulty adaptation.
• To develop and evaluate classification algorithms for inferring user states from physiological data.
• To demonstrate the feasibility of real-time, physiology-based difficulty adaptation in a competitive game.

Main Methods:

• An open-loop study involving 16 pairs playing a competitive game, measuring 5 physiological responses (respiration, skin conductance, ECG, 2 facial EMGs).
• Development of classification algorithms using physiological and performance data to predict self-reported variables (enjoyment, valence, arousal, perceived difficulty).
• A proof-of-concept closed-loop study to dynamically adapt game difficulty based on the developed classifiers.

Main Results:

• Highest classification accuracies were achieved for perceived difficulty (84.3% for 2-class, 60.5% for 3-class).
• The study demonstrated the technical feasibility of real-time, physiology-based difficulty adaptation.
• The closed-loop study did not find clear advantages for physiology-based adaptation over other methods.

Conclusions:

• Physiology-based affect recognition is a viable method for inferring user states in interactive scenarios.
• Real-time adaptation of task difficulty based on physiological linkage is technically feasible.
• Further research can leverage this approach to enhance user experience in various multi-user settings.