A Robust Strategy for Combining Several Classifiers for Small Samples and Heterogeneous Predictors

Charles Gomes (1) , Hisham Nocairi (1) , Marie Thomas (1) , Jean-François Collin (1) , Gilbert Saporta (2)

1 L'Oréal Recherche - L'Oréal Recherche France
2 CEDRIC - MSDMA - CEDRIC. Méthodes statistiques de data-mining et apprentissage

Faced to safety constraints, one cannot rely on a single prediction method, especially when the sample size is low. Stacking introduced by Wolpert (1992) and Breiman (1996) is a successful way of combining several models. We modify the usual stacking methodology when the response is binary and predictions highly correlated, by combining predictions with PLS-Discriminant Analysis instead of ordinary least squares. A strategy based on repeated split samples is then developed to select relevant variables and ensure the robustness of the final model. This method is applied to the prediction of hazard of 165 chemicals, based upon 35 in vitro and in silico characteristics.

Mots clés: supervised classification; stacking; Boosting; SVM; Safety evaluation; PLS-DA; Naïve Bayes

Mots clés: Évaluation de l'innocuité; Naïve Bayes; Stimulant; classification supervisée; empilage; PLS-DA

Domaines: Informatique [cs]; Statistiques [math.ST]

Charles Gomes, Hisham Nocairi, Marie Thomas, Jean-François Collin, Gilbert Saporta. A Robust Strategy for Combining Several Classifiers for Small Samples and Heterogeneous Predictors. Chimiometrie 2013, Sep 2013, Brest, France. pp.33-34. ⟨hal-01126303⟩