Improving Random Forests by Correlation-Enhancing Projections and Sample-Based Sparse Discriminant Selection
2016 (English)In: Proceedings 13th Conference on Computer and Robot Vision CRV 2016, Institute of Electrical and Electronics Engineers (IEEE), 2016, 222-227 p.Conference paper (Refereed)
Random Forests (RF) is a learning techniquewith very low run-time complexity. It has found a nicheapplication in situations where input data is low-dimensionaland computational performance is paramount. We wish tomake RFs more useful for high dimensional problems, andto this end, we propose two extensions to RFs: Firstly, afeature selection mechanism called correlation-enhancing pro-jections, and secondly sparse discriminant selection schemes forbetter accuracy and faster training. We evaluate the proposedextensions by performing age and gender estimation on theMORPH-II dataset, and demonstrate near-equal or improvedestimation performance when using these extensions despite aseventy-fold reduction in the number of data dimensions.
Place, publisher, year, edition, pages
Institute of Electrical and Electronics Engineers (IEEE), 2016. 222-227 p.
Correlation; Distributed databases; Estimation; Optimization; Radio frequency; Training; Vegetation; image classification
Computer Vision and Robotics (Autonomous Systems) Computer Engineering Software Engineering Signal Processing
IdentifiersURN: urn:nbn:se:liu:diva-134157DOI: 10.1109/CRV.2016.20ISI: 000392125600030ISBN: 9781509024919 (electronic)ISBN: 9781509024926 (electronic)OAI: oai:DiVA.org:liu-134157DiVA: diva2:1068782
13th Conference on Computer and Robot Vision, Victoria, British Columbia, Canada, 1-3 June 2016