Feature Selection Analysis of Chewing Activity Based on Contactless Food Intake Detection

Authors

  • Nur Asmiza Selamat Universiti Kebangsaan Malaysia
  • Sawal Hamid Md Ali Universiti Kebangsaan Malaysia
  • Khairun Nisa' Minhad Xiamen University Malaysia
  • Jahariah Sampe Universiti Kebangsaan Malaysia

Keywords:

Chewing detection, feature selection, proximity sensor, temporalis muscle, food intake detection

Abstract

This paper presents the feature selection methods for chewing activity detection. Chewing detection typically used for food intake monitoring applications. The work aims to analyze the effect of implementing optimum feature selection that can improve the accuracy of the chewing detection.  The raw chewing data is collected using a proximity sensor. Pre-process procedures are implemented on the data using normalization and bandpass filters. The searching of a suitable combination of bandpass filter parameters such as lower cut-off frequency (Fc1) and steepness targeted for best accuracy was also included. The Fc1 was 0,5Hz, 1.0Hz and 1.2H, while the steepness varied from 0.75 to 0.9 with an interval of 0.5. By using the bandpass filter with the value of [1Hz, 5Hz] with a steepness of 0.8, the system’s accuracy improves by 1.2% compared to the previous work, which uses [0.5Hz, 5Hz] with a steepness of 0.85. The accuracy of using all 40 extracted features is 98.5%. Two feature selection methods based on feature domain and feature ranking are analyzed. The features domain gives an accuracy of 95.8% using 10 features of the time domain, while the combination of time domain and frequency domain gives an accuracy of 98% with 13 features. Three feature ranking methods were used in this paper: minimum redundancy maximum relevance (MRMR), t-Test, and receiver operating characteristic (ROC). The analysis of the feature ranking method has the accuracy of 98.2%, 85.8%, and 98% for MRMR, t-Test, and ROC with 10 features, respectively. While the accuracy of using 20 features is 98.3%, 97.9%, and 98.3% for MRMR, t-Test, and ROC, respectively. It can be concluded that the feature selection method helps to reduce the number of features while giving a good accuracy.

Downloads

Download data is not yet available.

Downloads

Published

19-05-2021

How to Cite

Selamat, N. A., Md Ali, . S. H. ., Minhad, K. N., & Sampe, J. . (2021). Feature Selection Analysis of Chewing Activity Based on Contactless Food Intake Detection. International Journal of Integrated Engineering, 13(5), 38–48. Retrieved from https://publisher.uthm.edu.my/ojs/index.php/ijie/article/view/8620