Section: New Results

Handling numeric and temporal data in a local search-based classification algorithm

Participants: J. Jacques, L. Jourdan, C. Dhaenens, M. Vandomme

MOCA-I [20] is a highly efficient classification algorithm, primarily designed for knowledge extraction on large-scale, real-life medical data. This algorithm has been first extended to deal with numeric data [58] , [46] , through the definition of a model for classification rules on numeric attributes. Several neighborhood operators have been proposed, and compared, as components of the overarching local search metaheuristic guiding the discovery and optimization of these rules. A new model has also been proposed to handle temporal data. This model allows for the inclusion of sequences of events in classification rules, in addition to non-temporal attributes, thus building more informative classifiers. This model, along with various optimizations in the local search process, has been favorably compared to the previous MOCA-I algorithm and other standard classification algorithms. It is now used on real hospital data in order to evaluate its performance in a real environment.