Dipartimento di Statistica e Metodi Quantitativi

Martedì 6 ottobre 2015 ore 12.30

U7/2104 (AULA ECONOMICA) II PIANO

Quantile-based classifiers  
with application to the classification of bioaerosols

CHRISTIAN HENNIG 

Senior Lecturer, Department of Statistical Science, University College London

Abstract

Quantile classifiers are generalisations of the median-based classifiers recently introduced by 

Hall et al. (2009). They work for potentially high-dimensional data, and are defined by classi-

fying an observation according to a sum of appropriately weighted component-wise distances of 

the components of the observation to the within-class quantiles. The optimal quantiles can be 

chosen by minimizing the misclassification error in the training sample. I will present some 

theory and simulations results demonstrating that quantile classifiers are very competitive. 

Quantile classifiers will also be applied to the detection of bioaerosol particles based on gaseous 

plasma electrochemistry (Sarantaridis et al., 2012).

As several other classifiers, quantile classifiers aggregate information coming from the indivi-

dual variables. This depends on selection and standardisation of the variables. I will discuss the-

se issues in a way that may hopefully be of more general statistical interest. The first aspect is 

the extraction of some meaningful features from the high-dimensional and quite redundant ori-

ginal form of the bioaerosol data. The second aspect is to understand standardisation as va-

riable weighting, including the option to refrain from standardisation in situations where the 

variability of variables may be roughly proportional to the variable importance for classification, 

or to standardise certain groups of variables against each other, as will be suggested for the 

bioaerosol data.

Tutti gli interessati sono invitati a partecipare

per maggiori informazioni:

www.dismeq.unimib.it - francesca.greselin@unimib.it