Evaluation of Artificial Neural Network in Classifying Human Gender Based on Odour

Abstract

Biometrics is an advanced way of person recognition as it establishes more direct and explicit link with humans than passwords, since biometrics use measurable physiological and behavioral features of a person. In this paper, a gender recognition framework is proposed based on human odour. 20 samples of human odour from male and female are collected and only 16 out of 198 Volatile Organic Compounds (VOCs) are selected using the Chi-square test and entropy for gender detection and classification using artificial neural networks. In this paper, several different neural network activation functions were tested (e.g., Levenberg-Marquardt backpropagation, Gradient descent backpropagation and Resilient backpropagation) and several different neural network topologies are also tested with variety of hidden layers and number of neurons. It is also found that with 2 hidden layers having more number of neurons in the hidden layers (16 and 16 neurons in which hidden layer) was able to produce greater performance accuracy. The best learning algorithm that can be applied in gender detection shown in paper is the Gradient Descent learning algorithm. Also, it is notable that 8 out of 9 cases where all male samples are able to be detected or classified correctly compared to the 3 out of 9 cases in which all females are correctly detected or classified.