Session MA5 - Arbres, analyse de regroupement et méthodes basées sur les rangs pour problèmes complexes / Tree, Clustering and Rank Based Methods for Complex Problems
Day Monday, May 04, 2009 Room Ordre des CGA President Denis Larocque
Presentations
|
10h30 AM- 10h55 AM |
Mixed Effects Regression Trees for Clustered Data |
|
Ahlem Hajjem, HEC Montreal, Management Sciences We propose a new regression tree method that can handle unbalanced clusters, allows observations within a cluster to be in different nodes, and can incorporate random effects and observation-level covariates. Simulation results show that it can provide substantial improvements over standard tree algorithms. |
|
10h55 AM- 11h20 AM |
Trees and Ensemble Methods Applied to Multiple Outcomes with Mixture of Categorical and Continuous Responses |
|
Abdessamad Dine, HEC Montréal We proposed in the past a multivariate tree-based method for multiple outcomes with a mixture of categorical and continuous responses. In this presentation we compare the predictive performance of random forests built using the multivariate tree method and random forests built for each outcome individually. An illustration of the application of this new method is also given. |
|
11h20 AM- 11h45 AM |
Élaboration d’une nouvelle méthode de classification avec informations partielles sur la variable cible |
|
Tarek Grichi, HEC Montreal, Management Sciences La situation est la suivante. Nous disposons d’un échantillon où la valeur de la variable cible Y est connue pour uniquement une partie des observations. Plus précisément, la variable Y est binaire et seulement la classe ‘1’ est observée. Le but est de scorer la valeur de Y des autres observations de l’échantillon. |
|
11h45 AM- 12h10 PM |
Multivariate Nonparametric Methods for Clustered Data |
|
Denis Larocque, GERAD, HEC Montréal, Méthodes quantitatives de gestion, Canada During the last decade, there has been a wide interest to extend univariate and multivariate nonparametric procedures based on signs and ranks to clustered and hierarchical data. In this talk, I will present a review of these developments but the main focus will be on the more recent developments using the spatial signs and ranks. |