Incremental algorithm for decision rule generation in data stream contexts
- Autores: Nuria Mollá Campello
- Directores de la Tesis: Alejandro Rabasa Dolado (dir. tes.)
, Antonio Luis Ferrándiz Colmeiro (codir. tes.) , Joaquín Sánchez Soriano (codir. tes.) - Lectura: En la Universidad Miguel Hernández de Elche ( España ) en 2023
- Idioma: español
- Tribunal Calificador de la Tesis: Agustín Pérez Martín (presid.) , Rafael Molina Carmona (secret.) , Mary Daly (voc.)
- Enlaces
- Tesis en acceso abierto en: RediUMH
- Resumen
Nowadays, data science is earning a lot of attention in many different sectors. Specifically in the industry, many applications might be considered. Using data science techniques in the decision-making process is a valuable approach among the mentioned applications. Along with this, the growth of data availability and the appearance of continuous data flows in the form of data stream arise other challenges when dealing with changing data. This work presents a novel proposal of an algorithm, Incremental Decision Rules Algorithm (IDRA), that incrementally generates and modify decision rules for data stream contexts to incorporate the changes that could appear over time. This method aims to propose new rule structures that improve the decision-making process by providing a descriptive and transparent base of knowledge that could be integrated in a decision tool. This work describes the logic underneath IDRA, in all its versions, and proposes a variety of experiments to compare them with a classical method (CREA) and an adaptive method (VFDR). Some real datasets, together with some simulated scenarios with different error types and rates are used to compare these algorithms. The study proved that IDRA, specifically the reactive version of IDRA (RIDRA), improves the accuracies of VFDR and CREA in all the studied scenarios, both real and simulated, in exchange of more time.
