4 Presentations

• 
  01:30 PM - 01:55 PM

  The Markov Chain Importance Sampling Method for Rare-Event Probability Estimation

  • Zdravko Botev, presenter, The University of Montreal
  • Pierre L'Ecuyer, Université de Montréal

  We present a new Monte Carlo simulation based algorithm for the estimation of the probability of a (static) network failure. We argue that the proposed algorithm is competitive to existing procedures and is applicable to a broader spectrum of problems.

• 
  01:55 PM - 02:20 PM

  Randomly-Shifted Lattice Rules Adapted to Specific Integrands

  • David Munger, presenter, Université de Montreal
  • Pierre L'Ecuyer, Université de Montréal

  Randomly-shifted lattice rules, a variance reduction method for the simulation of stochastic estimators, replace independent Monte Carlo points with structured points more evenly distributed over the domain of integration. We compare approaches for constructing lattice rules adapted to an integrand in terms of the relative importance of the different projections.

• 
  02:20 PM - 02:45 PM

  Multivariate Forests with Missing Mixed Outcomes

  • Abdessamad Dine, presenter, HEC Montréal
  • Denis Larocque, GERAD - HEC Montréal
  • François Bellavance, GERAD - HEC Montréal

  In this paper, we propose a multivariate random forest method for multiple responses of mixed types with missing responses. Imputation is performed for each bootstrap sample used to build the individual trees that form the forest. The individual trees are built using a weighted splitting rule allowing downweighting of imputed observations. A simulation study shows the benefits of this approach over complete case analysis when missing responses are MCAR and MAR. In particular, the gain in prediction accuracy of the proposed method is larger in the MAR case and also increases as the proportion of missing increases.

• 
  02:45 PM - 03:10 PM

  Robustness of Random Forests for Regression

  • Marie-Helene Roy, presenter, HEC Montreal
  • Denis Larocque, GERAD - HEC Montréal

  In this study, we empirically investigate the robustness of random forests for regression problems. We also investigate the performance of five simple variations of the original random forest method, all aimed at improving robustness. Our results show that two of these variations offer good and stable performances.