An Intelligent Clinical Decision Support System for Patient-Specific Predictions to Improve Cervical Intraepithelial Neoplasia Detection

Table 3

Classification accuracies on the training, validation, and test sets of the 6 classifiers developed (single classifier approach).

Classifier	k-NN	NB	CART	MLP	RBF	PNN
Optimal parameters	k = 5	—	(i) Pruning level = 7/11; (ii) Number of terminal nodes = 8	2 hidden layers 18 × 18	(i) 486 neurons in the hidden layer (all samples of training set); (ii) Sigma = 0.6	Sigma = 0.4
Training set	78.6%	76.8%	77.6%	78.4%	87.6%	87.6%
Validation set	79.4%	80.9%	78.6%	77.0%	77.7%	80.2%
Test set	82.8%	82.8%	75.8%	80.5%	70.3%	80.0%

k-NN: k-nearest neighbours classifier, NB: naïve Bayesian classifier, CART: classification and regression tree, MLP: multilayer perceptron network, RBF: radial basis function network, PNN: probabilistic neural network.