On the capability of accommodating new classes within probabilistic neural networks

Abstract

To date, probabilistic neural networks (PNNs) have been widely used in various pattern classification tasks due to their robustness. In this paper, it is shown that by exploiting the flexible network configuration property, the PNN classifiers also exhibit the capability in accommodating new classes. This is verified by extensive simulation studies on using four different domain data sets for pattern classification tasks.

Electronic nose for black tea quality evaluation by an incremental RBF network

“…Another performance criterion of the classifier is the deterioration rate (dr) [22]. When new classes are added gradually in an incremental classifier, its performance degrades and dr gives a quantitative measure of the rate of degradation.…”

Comparison of customer response models

“…Due to the ease of training and a sound statistical foundation in Bayesian estimation theory, PNN has become an effective tool for solving many classification problems (Jin et al 2001;Hoya 2003).…”

“…Training a PNN actually consists mostly of copying training cases into the network, and so is as close to instantaneous as can be expected (Hoya 2003).…”

Enhanced probabilistic neural network with local decision circles: A robust classifier

“…Advantages of PPN are a) simplicity for not requiring a separate training phase for tuning network weights based on some learning rule and b) efficiency [35,93]. As such in recent years PNN has been used in a large number of applications such as seismic engineering [13], hydrology [86],ocean engineering [50],image processing [65,82], EEG, EKG, and other biomedical data analysis [58,[88][89][90]96], power engineering [28,57,73,87], and transportation engineering [61].…”