Identifying Novel Features from Specimen Data for the Prediction of Valuable Collection Trips.

Nicolson, Nicky; Tucker, Allan

Identifying Novel Features from Specimen Data for the Prediction of Valuable Collection Trips.

Resource type

Conference paper (published)

Creator (person)

Nicolson, Nicky

ORCID

Tucker, Allan

ORCID

Date published

October 2017

Abstract

Primary biodiversity data provide “what, where, and when” data points: the assertion that a species occurred at a particular point in space and time. These are most valuable when associated with specimens stored in natural history museums and herbaria, which evidence the assertions with reference to a physical specimen. The research presented uses novel data-mining techniques to uncover two hidden dimensions in specimen data - who collected the specimens and how they were collected. A combination of unsupervised and supervised learning techniques are used, which establish two new entities: collector and collection trip. Features are defined against these higher order representations of the data, which support the use of the data to answer novel questions such as which collection trips discover the most new species? We explore the features by building classifiers to predict species discovery, and compare these with a baseline model grouped using collector team transcriptions derived from the raw specimen data. Preliminary results are promising and whilst the particular focus of this research was botanical specimens, the technique is equally applicable to datasets of field-collected specimens from other scientific domains.