CLASSIFICATION OF FEATURE SELECTION BASED ON ARTIFICIAL NEURAL NETWORK

Hussain, Aini (2009) CLASSIFICATION OF FEATURE SELECTION BASED ON ARTIFICIAL NEURAL NETWORK. Industrial Electronic Seminar.

Preview

PDF Download (254Kb) | Preview

Abstract

Pattern recognition (PR) is the central in a variety of engineering applications. For this reason, it is indeed vital to develop efficient pattern recognition systems that facilitate decision making automatically and reliably. In this study, the implementation of PR system based on computational intelligence approach namely artificial neural network (ANN) is performed subsequent to selection of the best feature vectors. A framework to determine the best eigenvectors which we named as ‘eigenpostures’ of four main human postures specifically, standing, squatting/sitting, bending and lying based on the rules of thumb of Principal Component Analysis (PCA) has been developed. Accordingly, all three rules of PCA namely the KG-rule, Cumulative Variance and the Scree test suggest retaining only 35 main principal component or ‘eigenpostures’. Next, these ‘eigenpostures’ are statistically analyzed via Analysis of Variance (ANOVA) prior to classification. Thus, the most relevant component of the selected eigenpostures can be determined. Both categories of ‘eigenpostures’ prior to ANOVA as well as after ANOVA served as inputs to the ANN classifier to verify the effectiveness of feature selection based on statistical analysis. Results attained confirmed that the statistical analysis has enabled us to perform effectively the selection of eigenpostures for classification of four types of human postures.

Actions (login required)