SVM Applied to Trace Element Analysis of Human Hair

2 SVM Applied to Trace Element Analysis of Human Hair 

It has been recognized in medical science that the trace element contents in the hair of hypertension patients are different from those of normal persons. But in most of the published papers the data of hair analysis are treated simply by averaging or by correlation coefficient calculation. To use those simple methods only will give rise to significant loss of meaningful information contained in these data. So it is desirable to use more sophisticated algorithm to do this work. As an example, SVM has been employed to find the criterion of the hair samples of hypertension patients [33]. 

A total of 53 hair samples, including 26 samples from hypertension patients and 27 samples from normal persons, have been used for trace element analysis. The data of Ca, Mg, Al, Cu and Zn contents are listed in Table 10.1. 

It can be seen from Table 10.1 that the relationship between trace element contents and the hypertension seems not very clear if the content of any single trace element is considered, but some regularities can be revealed by multivariate analysis. By Fisher method, the rate of correct separation of the sample points of two classes is 83%, and the rate of correcmess of prediction by leave-one-out (LOO) cross-validation method is 81.1%. The rate of correct prediction by KNN method is 77.3%. 

By using Gaussian kernel, the data listed in Table 10.1 have been classified by SVM, the rate of correctness of separation is 96.25%, and that of prediction by LOO cross-validation method is 86.7%. It appears that the application of SVM can improve the research work about trace element contents in human hair. 

---