Analysis of Oral Cancer Prediction using Features Selection with Machine Learning

Accuracy is one of the main elements in the disease diagnose. Thus, it is important to select most relevant attributes to generate the optimal accuracy. The objective of this study is to predict more accurately the presence of oral cancer primary stage with reduced number of attributes. Originally, 25 attributes have been identified in order to predict the oral cancer staging. In this study, the integrated diagnostic model with hybrid features selection methods is used to determine the attributes that contribute the most to the diagnosis of oral cancer, which, indirectly, reduces the number of features that are collected from a variety of patient records. Twentyfive attributes have been reduced to 14 features using hybrid feature selection. Subsequently, four classifiers: Updatable Nave Bayes, Multilayer Perceptron, K-Nearest Neighbors and Support Vector Machine are used to predict the diagnosis of patients with oral cancer. Also, the observations indicate that the Support Vector Machine outperforms other machine learning algorithms after incorporating feature subset selection with SMOTE at preprocessing phases.