Neural Network and Decision Tree Induction: A Comparison in the Domain of Classification of Sonar Signal

Research into the problem of classification of sonar signals has been taken up as a challenging task for the neural networks. Our purpose is to make a comparison of decision tree induction and neural network classifiers in a classification of sonar signal databases. The task is to discriminate between sonar signals bounced off a metal cylinder and those bounced off a roughly cylindrical rock using discriminant functions such as multilayer perceptron neural network (MLP NN) and classification and regression tree (CART). This paper determines the optimal parameters of MLP NN model. It is shown that the proposed MLP NN classifier worked as an optimal classifier for the given task with an average classification accuracy at about 90.3609% on the test instances. The area under the receiver operating characteristics (ROC) curve for testing dataset is computed as 0.953, which is indicative of quality of the proposed classifier. The results also show that the proposed MLP NN classifier clearly outperforms the decision tree based classifiers on the testing datasets even after attempting different data partitions.