Principle Component Analysis Based Naïve Bayes and K-Nearest Neighbors Classifiers for the Identification of Oral Mucosal Lesions

Identification of the oral mucosal lesions has been executed by the machine learning methods in the present study. In the machine learning methods, principle component analysis, naïve Bayes and k-nearest neighbor’s classifiers, and receiver operating characteristic analysis are utilized. PCA is dimension reduction method and employed to reduce the dimension of the fluorescence data. Naïve Bayes is a probabilistic model which is based on the Bayes theorem. KNN is based on distance metric such as Euclidean distances. These methods are adopted for classification among the diseased (cancerous) and non-diseased groups. However, ROC is implemented for the finding of diagnostic parameters like sensitivity, specificity, and accuracy. Data collection was done by the use of fluorescence based portable device. A total of 38 patients (20 OSCC & 18 dysplastic), and 36 healthy volunteers were included in this study. Fluorescence spectral data were recorded from the lateral boarder of tongue (LBT) of the oral cavity. In the spectra, presence of two fluorophores namely FAD and porphyrin were seen. A significant enhancement in the porphyrin fluorescence was noticed on OSCC patients than the volunteers. Data analysis exhibited by the naïve Bayes and KNN classifiers on the testing data differentiated among Normal to OSCC, Normal to Dysplasia, and Dysplasia to OSCC with the accuracy values of 89%, 90%, 88% and 94%, 85%, and 100% respectively. Results indicate that KNN classifier can differentiate the three groups with slightly higher values of accuracy than the naïve Bayes classifier.