Cascade Correlation Neural Network Model for Classification of Oral Cancer

摘要

The rationale of this study is to accurately classify the records of the oral cancer patient on the basis of clinical symptoms, Gross Examination, Predisposing Factor, Histopathology, various tests and treatments. In this paper, Cascade correlation neural network model has been built as it combines together the idea of cascade architecture and learning algorithm together and it is estimated to be at least 10 times faster than standard back-propagation algorithms. The records of 1025 patients described with the help of 35 attributes are analysed to predict the rate of survivability of oral cancer patients. Dataset is divided in two subgroups: training subgroup and test subgroup, in order to verify the network's ability to diagnose new cases. Performance of the model for its ability to predict is evaluated on the basis of various measures. Classification accuracy of the model is 72.10%, sensitivity is 83.05%, specificity is 64.71%, precision of the model is 61.36%, recall capacity is 83.05%, f-measure value is 0.7058 and area under ROC curve is 0.944. Lift and Gain chart also suggest that cascade correlation neural network is an effective model for predicting oral cancer.