Rapilly, Mr Quentin (2020) Optimal Classification Trees PRE - Research Project, ENSTA.



The following document describes an alternative method to usual heuristic ones to build clas- sification trees. It consists in transforming the original classification problem in a Mixed Integer Optimisation (MIO) formulation. This has two main advantages : the first is that many powerful programs are dedicated to solve this kind of problem and do it very efficiently ; the second is that it allows to build the entire tree in one big step rather than building nodes recursively step by step. In a first part, the method, initially depicted by Dimitri Bertisimas and Jack Dunn, is summa- rized. The following parts are dedicated to its implementation some results and their analysis. I eventually present the results of some methods that I implemented in order to improve the original one.

Item Type:Thesis (PRE - Research Project)
Subjects:Mathematics and Applications
ID Code:8227
Deposited By:Quentin RAPILLY
Deposited On:19 mai 2021 15:28
Dernière modification:19 mai 2021 15:28

Repository Staff Only: item control page