Extraction de Règles d'Association Quantitatives - Application à des Données Médicales
Résumé
Mining association rules in databases has long been studied. However, most researches have focused on mining efficiently such rules in databases composed of boolean or categorical attributes, when in practice many tables contain also numeric attributes. In this paper, we propose QuantMiner, a system for mining multi-dimensional quantitative association rules. QuantMiner looks for the best intervals for numeric attributes relying on a genetic-based algorithm. Basically, in order to get high quality rules, both the support and confidence are optimized during the mining process. We conducted an intensive experimental evaluation of our algorithm on real datasets. Our experiments showed the usefulness of QuantMiner as an interactive data mining tool.