Visualization of anomaly data using peculiarity detection on learning vector quantization

Abstract

The purpose of this research is to develop the control chart robust for complex multidimensional data. In this study, we propose the methodology of anomaly data visualization and detection using hybrid model of Learning Vector Quantization (LVQ) and Peculiarity Factor (PF). LVQ is neural network model which uses supervised learning algorithm. It is useful to classification of multidimensional data with nonlinearity and multi-collinearity. PF is a criterion for evaluating peculiarity and is widely used for outlier detection. In the proposing method, PF of input data is calculated using the weight vector of LVQ. The anomaly data assigned to the class of the normal data was able to be displayed as an outlier on the control chart by calculation of PF on LVQ. The proposed model realized the robust discernment and visualization of the anomaly data that have complex distribution by small computational complexity. © 2013 Springer-Verlag Berlin Heidelberg.