Improved distance functions for the ReliefF family

This thesis continues from the preceding work from Robnik and Kononenko, where they demonstrate the importance and applicability of the Relief family of algorithms. This new dissertation emphasizes the augmentation of input data (numerical, categorical, and text data types) that the Relief algorithm...

Descripción completa

Detalles Bibliográficos
Autor: Moran Calderon, Augusto
Tipo de recurso: tesis de maestría
Fecha de publicación:2023
País:España
Institución:Universitat Politècnica de Catalunya (UPC)
Repositorio:UPCommons. Portal del coneixement obert de la UPC
Idioma:inglés
OAI Identifier:oai:upcommons.upc.edu:2117/407783
Acceso en línea:https://hdl.handle.net/2117/407783
Access Level:acceso abierto
Palabra clave:Algorithms
Embeddings (Mathematics)
Relief
ReliefF
RReliefF
feature selection
feature estimation
quality estimates
SBERT
embeddings
distance functions
Algorismes
Embeddings (Matemàtica)
Àrees temàtiques de la UPC::Informàtica::Informàtica teòrica::Algorísmica i teoria de la complexitat
Descripción
Sumario:This thesis continues from the preceding work from Robnik and Kononenko, where they demonstrate the importance and applicability of the Relief family of algorithms. This new dissertation emphasizes the augmentation of input data (numerical, categorical, and text data types) that the Relief algorithm can wield for two types of problems (classification and regression) while preserving relevant properties such as robustness and flexibility. Moreover, we provide an exhaustive in-depth analysis of the algorithm convergence. Stating how it works, and where are the major points of failure associated with applying the algorithm to real-world datasets. As well as valid arguments guided by well-known research experiments to reinforce our implementation choices. In addition to an empirical analysis with synthetic and real datasets. An available software framework is made public using open-source libraries.