Formulation of sensor ranking associated in categorical perception: a pilot study using machine learning

• Main Authors: Bakar, Abdul Rauf A., Lai, Khin Wee, Hamzaid, Nur Azah
• Format: Conference or Workshop Item
• Published: Springer Science and Business Media Deutschland GmbH 2022
• Subjects: TA Engineering (General). Civil engineering (General)
• Online Access: http://eprints.um.edu.my/43404/; https://www.scopus.com/inward/record.uri?eid=2-s2.0-85129321333&doi=10.1007%2f978-3-030-90724-2_1&partnerID=40&md5=c9d95432d0832a41454066d8d1040816

The concept of categorical perception has been enormously investigated to digitalize the process of auditory processing when perceiving speech stimulus at higher brain neurological signal. Despite the nature non-stationary property of electroencephalography (EEG) during any task processing, scientists and clinicians find it to be not well fitted in the healthcare setting application without professional monitoring, and even worst, the hearing aid device functionality was also reported at a low success rate. In this study, we embraced machine learning technology and extracting contribution into our auditory research area. The present pilot work aims to create a robust computational framework to formulate the sensor ranking principle in auditory speech perception. The ranking for sensors could facilitate in identifying the minimal set sensor-of-interest (ROI) that are sufficient in specific auditory task processing using an optimally trained model. The trained Support Vector Machine (SVM) highest performance reported at random 2 training dataset with scoring of 92.3 using 70 triple-random training dataset. Based on the sensor ranking, the CZ electrode outperformed the other electrodes with scoring of 96.74, followed by PZ and FPZ for the 2nd and 3rd rank (95.66 and 95.34 respectively). Our pilot study anticipated that the sensor ranking formula able to underline more precise neural correlates based on current auditory categorical perception response. The excellence sensor ranking in delivering a minimal set of sensor-of-interest (SOI) drive the capability of the SVM model in classifying auditory brain response in high-performance prediction metrics and possible reliability in the healthcare setting application. © 2022, Springer Nature Switzerland AG.