A dropout-regularised neural network for mapping arsenic enrichment in SW England using MXNet

Kirkwood, Charlie. 2016 A dropout-regularised neural network for mapping arsenic enrichment in SW England using MXNet. [Poster] In: British Geological Survey Poster Competition 2016, Nottingham, UK, Dec 2016. British Geological Survey. (Unpublished)

Before downloading, please read NORA policies.

Preview

Text (Poster) A dropout-regularised neural network for mapping arsenic enrichment in SW England using MXNet HQ.pdf Download (5MB) | Preview

Abstract/Summary

This poster applies a dropout-regularised artifical neural network, constructed in the MXNet framework, to map arsenic enrichment in south west England. The network models the relationships between arsenic (as a centred log-ratio from XRF analyses of 3395 stream sediment samples) and high resolution geophysical data. The resultant model, trained to optimal accuracy using early stopping, achieves an R2 of 0.7 on held-out test data - a promising level of accuracy for predictions in a complex hydrothermal mineralisation system, and warrants further investigation into the development of more sophisticated network architectures for geohazard and prospectivity mapping purposes.

Item Type:	Publication - Conference Item (Poster)
Additional Keywords:	Artificial Neural Networks, Machine learning, Artificial intelligence, Prospectivity mapping, Geochemistry, Geological mapping, Data Science, Data Analysis, Geology
NORA Subject Terms:	Earth Sciences Ecology and Environment Electronics, Engineering and Technology Computer Science
Date made live:	28 Sep 2016 12:34 +0 (UTC)
URI:	https://nora.nerc.ac.uk/id/eprint/514614

Actions (login required)

View Item

Document Downloads

More statistics for this item...