SYMBOLIC ONE-CLASS LEARNING FROM IMBALANCED DATASETS: APPLICATION IN MEDICAL DIAGNOSIS

• DOI: 10.1142/s0218213009000135
• Publication Date: 2009-04-27T13:26:44Z
• Authors: LUIS MENA; JESUS A. GONZALEZ
• Keywords: Abstraction

When working with real-world applications we often find imbalanced datasets, those for which there exists a majority class normal data and minority abnormal or important data. In this work, make an overview of the imbalance problem; review consequences, possible causes existing strategies to cope inconveniences associated problem. As effort contribute solution problem, propose new rule induction algorithm named Rule Extraction MEdical Diagnosis (REMED), as symbolic one-class learning approach. For evaluation proposed method, use different medical diagnosis datasets taking into account quantitative metrics, comprehensibility, reliability. We performed comparison REMED versus C4.5 RIPPER combined over-sampling cost-sensitive strategies. This empirical analysis showed it be quantitatively competitive in terms area under Receiver Operating Characteristic curve (AUC) geometric mean, but overcame them comprehensibility Results our experiments show that generated rules systems larger degree abstraction patterns closer well-known values each considered dataset.