ENBIS-17 in Naples

9 – 14 September 2017; Naples (Italy) Abstract submission: 21 November 2016 – 10 May 2017

Stacking Prediction for a Binary Outcome

11 September 2017, 10:50 – 11:10

Abstract

Submitted by
Hicham Noçairi
Authors
Hicham Noçairi (L’Oréal Research and Innovation), Charles Gomes (L’Oréal Research and Innovation), Marie Thomas (L’Oréal Research and Innovation), Gilbert Saporta (CEDRIC-CNAM)
Abstract
A large number of supervised classification models have been proposed in the literature. In order to avoid any bias induced by the use of one single statistical approach, they are often combined through a specific “stacking” meta-model. To deal with the case of a binary outcome and of categorical predictors, we introduce several improvements to stacking: combining models is done through PLS-DA instead of OLS due to the strong correlation between predictions, and a specific methodology is developed for the case of a small number of observations, using repeated sub-sampling for variable selection. Five very different models (Boosting, Naïve Bayes, SVM, Sparse PLS-DA and Expert Scoring) are combined through this improved stacking, and applied in the following context: according to the 7th Amendment of the European Cosmetic Directive, L’Oréal is committed to stop animal testing. Consequently, l’Oréal must develop alternative approaches for safety evaluation of chemicals (irritation, sensitization). In vitro, in silico and physicochemical parameters are used to predict the in vivo tests results (danger/non danger). Results show that stacking meta-models have better performances than each of the five models taken separately (a 5%-10% average increase) and furthermore, stacking provides a better balance between sensitivity and specificity.

Return to programme