A comparison of GE optimized neural networks and decision trees

  • Authors:

  • Kristopher Hoover;Rachel Marceau;Tyndall Harris;David Reif;Alison Motsinger-Reif

  • Affiliations:

  • NC State University Institute for Advanced Analytics, Raleigh, NC, USA;NC State University, Raleigh, NC, USA;NC State University, Raleigh, NC, USA;NC State Unversity, Raleigh, NC, USA;NC State University, Raleigh, NC, USA

  • Venue:
  • Proceedings of the 14th annual conference companion on Genetic and evolutionary computation
  • Year:
  • 2012

Quantified Score

Hi-index 0.00

Visualization

Abstract

Grammatical evolution neural networks (GENN) is a commonly utilized method at identifying difficult to detect gene-gene and gene-environment interactions. It has been shown to be an effective tool in the prediction of common diseases using single nucleotide polymorphisms (SNPs). However, GENN lacks interpretability because it is a black box model. Therefore, grammatical evolution of decision trees (GEDT) is being considered as an alternative, as decision trees are easily interpretable for clinicians. Previously, the most effective parameters for GEDT and GENN were found using parameter sweeps. Since GEDT is much more intuitive and easy to understand, it becomes important to compare its predictive power to that of GENN. We show that it is not as effective as GENN at detecting disease causing polymorphisms especially in more difficult to detect models, but this power trade off may be worth it for interpretability.