The improvement of the Learning Environment in the context of Multi-label data

Résumé

Over the last few years, Multi-label Learning (MLL) has attracted the attention of a large community of researchers in many fields. Initially, it was applied for text categorization in which the annotation of a document that belongs to multiple categories require specific approaches. Thereafter, MLL is being increasingly required in other many real-world applications. In our work, we considered MLL for the medical aid diagnosis, our first research goal was the investigation of the advantages of using committee of learners to improve a Multilabel algorithm that adapts K-Nearest-Neighbors (KNN) to Multi-label problem called MLKNN using Bagging and Boosting. - Secondly, we gathered a medical Multi-label dataset that concerns Ambulatory Blood Pressure Monitoring (ABPM) which currently occupies a central place in the diagnosis and follow-up of hypertensive patients. We also proposed, an intelligent analysis of ABPM records using Multi-label Classification algorithms allowing the expert to analyze them more quickly and efficiently. In addition, it could help to investigate label dependencies and provide interesting insights. The satisfactory findings and interpretations of this work, conducted us to investigate more about the advantages of using Decision Trees (DT) to extract new and implicit correlations between different labels and features in a given dataset. For that, we reviewed recent works addressing Label dependencies based on several Multi-label algorithms based on DT. We presented also the main differences between the two defined types of Label correlation named Conditional and Unconditional (Marginal) Label dependence. Finally, we conducted a comparative study of six well-known algorithms in the literature, and we discussed the benefits of considering Label dependence using DT algorithm as a base classifier for both Transformation and Adaptation algorithms. Finally, potential further works and future directions of our thesis were highlighted.

Citer ce document

Accès au document

Ce document est hébergé sur son dépôt institutionnel d'origine.

Auteur(s)

Statistiques

Consultations : 2

Téléchargements : 0