Optimal decision trees using optimization techniques

The rising need of having a way to understand and explain the decisions produced by the artificial intelligence algorithms, used in a broad set of fields, led to the apparition of the concept of explainable artificial intelligence. One of the most simple, although powerful, algorithms are the decisi...

Descripción completa

Detalles Bibliográficos
Autor: Alòs Pascual, Josep
Tipo de recurso: tesis de maestría
Fecha de publicación:2020
País:España
Institución:Universitat Oberta de Catalunya (UOC)
Repositorio:O2, repositorio institucional de la UOC
OAI Identifier:oai:openaccess.uoc.edu:10609/117686
Acceso en línea:http://hdl.handle.net/10609/117686
Access Level:acceso abierto
Palabra clave:arbres de decisió
business intelligence
optimització
arboles de decisión
optimización
decision trees
optimization
Data mining -- TFM
Mineria de dades -- TFM
Minería de datos -- TFM
Descripción
Sumario:The rising need of having a way to understand and explain the decisions produced by the artificial intelligence algorithms, used in a broad set of fields, led to the apparition of the concept of explainable artificial intelligence. One of the most simple, although powerful, algorithms are the decision trees. This project focuses on studying the algorithms that allow the creation of such trees, while ensuring that the tree is optimal, as smaller trees are usually easier to explain. The project presents a Python package whose purpose is to act as a barrier remover for the users that don't have the means to implement those algorithms, allowing them to use the implementations proposed by different authors while leveraging the implementation of both the algorithms and the interaction with the solving ls to the package. In this report, the design of such tool is presented, as well as the technical considerations on which solving tools are used. Also, benchmarking on different datasets used in the bibliography is done to assess that the package accomplishes its main task, and to compare the different approaches implemented.