A Method of Peculiarity Factor Approximation Using Vector Quantization Model

作者：Saitoh Fumiaki^*; Ishizu Syohei

来源：Electronics and Communications in Japan, 2016, 99(8): 63-71.

DOI：10.1002/ecj.11840

• vector quantization
• peculiarity factor (PF)
• competitive learning
• data mining anomaly detection
• computational reduction

The purpose of this research is to reduce the computational complexity of the peculiarity factor (PF). Recently, PF has been adopted as the index for anomaly data detection, and it is widely used in various mining scenes. The fact that PF has become a powerful mining tool is positive because its calculation method is extremely simple and the results of the calculation are easy to understand visually. One of the most important problems for using PF for large-scale data is the rapidly increasing computational complexity required when the data volume increases. The computational complexity of PF is in the polynomial order because the PF of each data is calculated distantly over all the data. In this study, we propose an approximation methodology for PF for computational reduction and for enhanced robustness using the vector quantization model. Approximate values of PF are calculated by replacing the actual data with the nodes of vector quantization model. By calculating PF based on the vector quantization node vectors, we achieve restraint in the increasing computational complexity.

• 出版日期2016-8