Analysis and Improvements of the Adaptive Discretization Intervals Knowledge Representation

In order to handle classification problems with real-valued attributes using discretization algorithms it is necessary to obtain a good and reduced set of cut points in order to learn successfully. In recent years a discretization-based knowledge representation called Adaptive Discretization Intervals has been developed that can use several discretizers at the same time and also combines adjacent cut points. In this paper we analyze its behavior in several aspects. From this analysis we propose some fixes and new operators that manage to improve the performance of the representation across a large set of domains.

• Type: Conference
• Year: 2004
• Where: GECCO
• Authors: Jaume Bacardit, Josep Maria Garrell i Guiu
• Keywords: Adaptive Discretization Intervals, Cut Points, Discretization Algorithms, GECCO 2004