Arguing from Experience to Classifying Noisy Data

A process, based on argumentation theory, is described for classifying very noisy data. More specifically a process founded on a concept called “arguing from experience” is described where by several software agents “argue” about the classification of a new example given individual “case bases” containing previously classified examples. Two “arguing from experience” protocols are described: PADUA which has been applied to binary classification problems and PISA which has been applied to multi-class problems. Evaluation of both PADUA and PISA indicates that they operate with equal effectiveness to other classification systems in the absence of noise. However, the systems out-perform comparable systems given very noisy data.

• Type: Conference
• Year: 2009
• Where: DAWAK
• Authors: Maya Wardeh, Frans Coenen, Trevor J. M. Bench-Capon
• Keywords: Argumentation Theory, Binary Classification Problems, DAWAK 2009, Noisy Data