New Self-Adaptive Probabilistic Neural Networks in Bioinformatic and Medical Tasks

Abstract

We propose a self–adaptive probabilistic neural network model, which incorporates optimization algorithms to determine its spread parameters. The performance of the proposed model is investigated on two protein localization problems, as well as on two medical diagnostic tasks. Experimental results are compared with that of feedforward neural networks and support vector machines. Different sampling techniques are used and statistical tests are conducted to calculate the statistical significance of the results.

Assessment of prediction ability for reduced probabilistic neural network in data classification problems

“…The best results are marked with bold. As shown, in each data classification case, the PNN models trained by means of Algorithm 1 or Algorithm 2 outperform PNNs trained using state-of-the-art methods (Georgiou et al 2006;Chang et al 2008;Georgiou et al 2008;Saiti et al 2009;Temurtas et al 2009;Chandra and Babu 2011;Yeh and Lin 2011;Azar and El-Said 2013). Our algorithms also perform better than the reference classifiers in all considered data set classification problems.…”

Fuzzy Evolutionary Probabilistic Neural Networks

“…Moreover, the CPU training times are also reported in Tables 3 and 4. In order to evaluate the performance of our model, we have applied these six benchmark problems to Homoscedastic and Heteroscedastic Evolutionary Probabilistic Neural Networks [1] as well as to original PNNs and Bagging EPNNs [2]. For the original PNN's implementation, an exhaustive search for the selection of the spread parameter σ has been conducted in the interval [10 −3 , 5] and the σ that resulted to the best classification accuracy on the training set has been used for the calculation of PNN's classification accuracy on the test set.…”

“…The training procedure of a PNN is quite simple and requires only a single pass of the patterns of the training data which results to a short training time. The architecture of a PNN always consists of four layers: the input layer , the pattern layer , the summation layer and the output layer [1,3].…”

“…One way of obtaining this desirable property is to incorporate a Fuzzy Membership Function (FMF) into the model. Among the large variety of classification models we chose the EPNN due to its simplicity, effectiveness and efficiency [1,2] and we have incorporated the FMF proposed in [15] for weighting the pattern neurons of the EPNN. By this way, fuzzy class membership values are assigned to each pattern neuron and for this reason the proposed model is named Fuzzy Evolutionary Probabilistic Neural Network (FEPNN).…”

“…A simple but promising model which combines statistical methods and efficient evolutionary algorithms is the recently proposed Evolutionary Probabilistic Neural Network (EPNN) [1,2]. Specifically, EPNN is based on the Probabilistic Neural Network (PNN) introduced by Specht [3] that has been widely used in several areas of science with promising results [4,5,6,7].…”