MULTIPLE DISCRIMINANT ANALYSIS APPLIED TO COUGH CATEGORIZATION,
Abstract
The objective of this research is to evaluate the effectiveness of a multiple discriminant model for the detection of coughs and the use of factor analysis techniques to reduce the dimension of the measure set. Data for the training and testing of the model is in the form of analog recordings of coughs and artifacts (noncoughs). The recordings are preprocessed to normalize the recording levels and delete sections of data containing only low noise levels. After removal of the low level noise, continuous sections of digitized data contain either a cough or an artifact. The multi-variate linear discriminant model used in this research assumes that the measurements computed for each subsegment are normally distributed and the covariance matrices for the two categories of data are equal. The program includes a factor loading procedure for estimating the importance of each measure in the classification of the data. Also consideration is given to the problem of the selection of a small subset of the best measures. The accuracy of the classifier using the entire measure set is compared to its accuracy when a small group of the best measures is used. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 09, 1969
- Accession Number
- AD0694127
Entities
People
- A. J. Welch
Organizations
- University of Texas at Austin