| Summary: | Use of SERS spectra to detect NS1 in saliva is a most current finding that could lead to early, non-invasive, non-blood infectious detection of diseases related to NS1. Since the volume of the SERS spectral data is humongous, hence an automated analysis technique to classify the NS1 adulterated samples is vital to the success of this method. K-Nearest Neighbor (K-NN) is a popular classifier that searches the space for k training records that are nearest to the new record as the neighbors of new records. Our work here intends to find an optimal k-NN classifier for detecting NS1 adulterated salivary samples from the SERS spectra. A total of 128 spectra, each with 1801 Raman shifts were analyzed. The performance of the k-NN classifier, in terms of accuracy, precision, sensitivity and specificity, at different values of nearest neighbour (k) were investigated. Results show that the value of nearest neighbour for the optimal k-NN classifier is between 1 to 11. The optimal k-NN classifier is strictly specific, 100% for all k, and highly sensitive, 89.5% for k between 1 to 11. It has a precision of 100% for all k and an accuracy of 92.3% for k between 1 to 11. © 2016 IEEE.
|
|---|