Oral Cancer Prediction Using Gene Expression Profiling and Machine Learning

Oral premalignant lesion (OPL) patients have a high risk of developing oral cancer. In this study we investigate using machine learning techniques with gene expression profiling to predict the possibility of oral cancer development in OPL patients. Four classification techniques were used: support vector machine (SVM), Regularized Least Squares (RLS), multi-layer perceptron (MLP) with back propagation and deep neural network (DNN). Fisher discriminate analysis was used to select relevant features from the gene expression array. The results show high accuracy (96%) using DNN and 94% accuracy using SVM and MLP with one sample cross validation. Furthermore, we achieved the same results using 10-fold cross validation.