Variable Selection Using Adaptive Band Clustering and Physarum Network

• Main Authors: Huanyu Chen, Tong Chen, Zhihao Zhang, Guangyuan Liu
• Format: Article
• Language: English
• Published: MDPI AG 2017-06-01
• Series: Algorithms
• Subjects: affinity propagation; physarum network; variable selection; wavelength selection; real-time spectroscopy; on-line analysis
• Online Access: http://www.mdpi.com/1999-4893/10/3/73

Variable selection is a key step for eliminating redundant information in spectroscopy. Among various variable selection methods, the physarum network (PN) is a newly-introduced and efficient one. However, the whole spectrum has to be equally divided into sub-spectral bands in PN. These division criteria limit the selecting ability and prediction performance. In this paper, we transform the spectrum division problem into a clustering problem and solve the problem by using an affinity propagation (AP) algorithm, an adaptive clustering method, to find the optimized number of sub-spectral bands and the number of wavelengths in each sub-spectral band. Experimental results show that combining AP and PN together can achieve similar prediction accuracy with much less wavelength than what PN alone can achieve.