Section: New Results
MOCA-I: Multi-Objective Classification Algorithm for Imbalanced Data
Participants: Julie Jacques, Clarisse Dhaenens, Laetitia Jourdan
Dealing with Imbalanced data is a real challenge as predicting the minority class may be very difficult but has a great interest for medical applications for example. Therefore, we propose MOCA-I, a new multi-objective local search algorithm that is conceived to deal with class imbalancy, double meaning of missing data, volumetry and need of highly interpretable results all together  . MOCAI is a Pittsburgh multi-objective partial classification rule mining algorithm, using dominance-based multi-objective local search (DMLS). In comparison to state-of-the-art classification algorithms, MOCA-I obtains the best results on the 10 data sets of literature and is statistically better on the real data sets  .