Comparing Methods of Feature Extraction of Brain Activities for Octave Illusion Classification Using Machine Learning

Area of Science:

• Neuroscience
• Machine Learning
• Auditory Perception

Background:

• The octave illusion is a complex auditory phenomenon.
• Distinguishing between octave illusion and non-illusion states in brain activity is challenging.
• Automatic feature extraction from auditory steady-state responses (ASSR) is crucial for objective analysis.

Purpose of the Study:

• To identify efficient automatic feature selection methods for ASSR data.
• To compare the performance of different machine learning algorithms in classifying octave illusion.
• To improve the accuracy of distinguishing octave illusion from non-illusion groups based on brain activity.

Main Methods:

• Compared four automatic feature selection techniques: univariate selection, recursive feature elimination, principal component analysis, and feature importance.
• Utilized machine learning algorithms including linear regression, random forest, and support vector machine (SVM).
• Evaluated methods based on their accuracy in classifying auditory octave illusion and non-illusion groups using ASSR amplitude differences.

Main Results:

• Univariate selection combined with SVM classification yielded the highest accuracy at 75%.
• This represents a significant improvement over the 66.6% accuracy achieved without feature selection.
• The study demonstrated the effectiveness of specific feature selection methods in enhancing classification performance.

Conclusions:

• Univariate feature selection with SVM is an effective method for classifying octave illusion based on ASSR data.
• The findings provide a foundation for future research into the mechanisms of the octave illusion.
• This work paves the way for developing algorithms for automatic octave illusion classification.