Machine learning identifies the dynamics and influencing factors in an auditory category learning experiment

Area of Science:

• Psychology
• Cognitive Neuroscience
• Machine Learning

Background:

• Analyzing human learning data is complex due to confounding factors like perception, feedback, and individual differences.
• Auditory category learning presents unique challenges in modeling participant progress and performance saturation.

Purpose of the Study:

• Investigate human learning evolution in auditory category tasks using machine learning.
• Determine how participant performance saturates and identify early indicators of learning.
• Assess the feasibility of differentiating learned categories before the experiment concludes.

Main Methods:

• Applied Gaussian Mixture Models to analyze participant performance evolution and identify task configuration influences.
• Utilized CatBoost, a classification algorithm, to distinguish between participants who learned categories and those who did not.
• Analyzed data from an auditory category-learning experiment.

Main Results:

• Gaussian Mixture Models effectively described performance trends and informed factor identification (Q1).
• Observed early saturation trends in participant performance (Q2).
• CatBoost accurately separated learners from non-learners significantly before the session's end (Q3).

Conclusions:

• Machine learning provides powerful tools for modeling participant dynamics in learning tasks.
• Identified key factors influencing task design and performance.
• Results support improved computational models and suggest optimal timing for learning interventions.