A new algorithm for redundancy minimisation in geo-environmental data

被引：6

作者：

Laib, Mohamed ^{[1
]}

Kanevski, Mikhail ^{[1
]}

机构：

[1] Univ Lausanne, Fac Geosci & Environm, Inst Earth Surface Dynam, CH-1015 Lausanne, Switzerland

来源：

COMPUTERS & GEOSCIENCES | 2019年 / 133卷

关键词：

Redundancy minimisation; Unsupervised feature selection; Space filling; Geo-environmental data; Machine learning; UNSUPERVISED FEATURE-SELECTION; MORISITA ESTIMATOR; UNIFORM DESIGNS; DIMENSION;

D O I：

10.1016/j.cageo.2019.104328

中图分类号：

TP39 [计算机的应用];

学科分类号：

081203 ; 0835 ;

摘要：

Identification of the relevant features in modelling of spatial environmental phenomena is an ongoing challenge in data driven modelling. The research proposes a new unsupervised learning algorithm to reduce the redundancy in the input space. The algorithm is based on the adaptation of a space filling criterion called the coverage measure. This measure has been used to judge the quality of an experimental design and select the best set of points. In the present work the goal is to select the most useful subset of the independent variables (features of the input space) in a dataset by reducing the existing redundancy. The main advantage of this algorithm is its easiness of implementation and it is parameter-free approach. The paper is accompanied with a new R package "SFtools" that implements the algorithm.

引用

页数：9