PCA-KNN for Detection of NS1 from SERS Salivary Spectra

• Published in: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
• Main Author: Othman N.H.; Lee K.Y.; Radzol A.R.M.; Mansor W.; Wong P.S.; Looi I.
• Format: Conference paper
• Language: English
• Published: Springer Verlag 2018
• Online Access: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85043576548&doi=10.1007%2f978-3-319-75420-8_32&partnerID=40&md5=9c9e5c988b028bc3bcea81f9e58506ec
• ISSN: 3029743
• DOI: 10.1007/978-3-319-75420-8_32

K-Nearest Neighbor (kNN) has shown its strong capability in pattern recognition, classification and machine learning applications. In this paper, kNN was used to distinguish between Non-structural protein 1 (NS1) positive and NS1 negative dengue patients from salivary Raman spectra. The presence of NS1 was detected in the saliva of dengue infected subjects. It was found Raman active, producing a molecular Raman fingerprint. Surface Enhanced Raman Spectroscopic (SERS) technique was adopted in obtaining the NS1 Raman spectra dataset. Performance of kNN with different K-values, optimized with Scree, Cumulative Percentage Variance (CPV) and Eigenvalue One Criterion (EOC) stopping criteria, was investigated and compared in term of sensitivity, specificity, accuracy and kappa. The best performance is found with the use of CPV stopping criteria and a K-value of 5, which attained an accuracy of 84.5% and kappa of 0.69. © Springer International Publishing AG, part of Springer Nature 2018.