Extraction of decision rules via imprecise probabilities
Data analysis techniques can be applied to discover important relations among features. This is the main objective of the Information Root Node Variation (IRNV) technique, a new method to extract knowledge from data via decision trees. The decision trees used by the original method were built using...
| Autores: | , , , |
|---|---|
| Formato: | artículo |
| Fecha de publicación: | 2017 |
| País: | España |
| Recursos: | Universitat Politècnica de València (UPV) |
| Repositorio: | RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia |
| Idioma: | inglés |
| OAI Identifier: | oai:riunet.upv.es:10251/149156 |
| Acesso em linha: | https://riunet.upv.es/handle/10251/149156 |
| Access Level: | acceso abierto |
| Palavra-chave: | Imprecise probabilities Imprecise Dirichlet model Non-parametric predictive inference model Uncertainty measures Decision rules Traffic accident severity INGENIERIA E INFRAESTRUCTURA DE LOS TRANSPORTES |
| Resumo: | Data analysis techniques can be applied to discover important relations among features. This is the main objective of the Information Root Node Variation (IRNV) technique, a new method to extract knowledge from data via decision trees. The decision trees used by the original method were built using classic split criteria. The performance of new split criteria based on imprecise probabilities and uncertainty measures, called credal split criteria, differs significantly from the performance obtained using the classic criteria. This paper extends the IRNV method using two credal split criteria: one based on a mathematical parametric model, and other one based on a non-parametric model. The performance of the method is analyzed using a case study of traffic accident data to identify patterns related to the severity of an accident. We found that a larger number of rules is generated, significantly supplementing the information obtained using the classic split criteria. |
|---|