Wavelet Entropy Based Probabilistic Neural Network for Classification

Abstract

Recently, wavelet transform (WT) has been enormously effectual in various scientific fields. As a matter of fact, WT has overcome the FFT in the difficult nature data tackling. A wavelet entropy based probabilistic neural network (PNN) for classification applications is proposed. Specifically, wavelet transform is performed on the original input feature data, and the entropy values of the wavelet decomposition signals are then extracted to use as the input to the PNN classifier. Two benchmark data sets, Breast Cancer and Diabetes, are used to demonstrate the efficiency of our proposed wavelet entropy based PNN (WEPNN) classifier. The test classification rates of 80.3% and 77.0% are achieved respectively for the two data sets using the WEPNN with Shannon entropy. Other published methods are used for comparison. The method is promising. For results accuracy enhancement, large data set might be utilized in the future work.

Classification of Metastatic and Non-Metastatic Thoracic Lymph Nodes in Lung Cancer Patients Based on Dielectric Properties Using Adaptive Probabilistic Neural Networks

“…Probabilistic Neural Network Model PNN was used in this study (6)(7)(8)(9)(10). PNN is composed of four layers: input layer, radial base layer (pattern layer), decisionmaking layer (summation layer) and output layer ( Figure 1).…”

“…Probabilistic neural network (PNN) was first proposed by Donald F. Specht in the late 18 th century. The theoretical basis of the network is Bayesian classification theory and probability density function estimation (6)(7)(8)(9)(10). It can realize the function of nonlinear learning algorithms with linear learning algorithms, which is widely applied in pattern classification problems.…”