Sampled clustering on phenotypic data of plant

Abstract

Clustering species of the same plant into di↵erent groups is an important step in developing new species of the concerned plant. Phenotypic (or Physical) characteristics of plant species are commonly used to perform clustering. Hierarchical Clustering (HC) is popularly used for this task, and this algorithm su↵ers from low accuracy and high computational complexity. In one of the recent works [1], the authors used the standard Spectral Clustering (SC) algorithm to improve the clustering accuracy. Further, they used Pivotal Sampling, a probabilistic algorithm, to reduce the cost of clustering. They demonstrated the efficacy of their algorithm on soybean species. Use of the standard SC algorithm is not ideal. Based upon spectral graph theory and the involved Cheeger’s inequality, we propose a novel base “a” SC instead. We further improve our algorithm by integrating the technique of local scaling (instead of global one) as proposed in [2]. We perform experiments on 1865 rice species. Using eigenvalue analysis, we argue that both our algorithms should lead to a better clustering. We further demonstrate that the standard SC is 49.86% better than HC while our base “a” SC and base “a” locally scaled SC are 64.93% and 66.33% better than HC, respectively. We also demonstrate that our latter clustering algorithm is better than another popular clustering technique of Gaussian Mixture Model (22.05% better).