This document describes a novel learning algorithm that classifies "bags" ofinstances rather than individual instances. A bag is labeled positive if itcontains at least one positive instance (which may or may not be specificallyidentified), and negative otherwise. This class of problems is known asmulti-instance learning problems, and is useful in situations where the classlabel at an instance level may be unavailable or imprecise or difficult toobtain, or in situations where the problem is naturally posed as one ofclassifying instance groups. The algorithm described here is an ensemble-basedmethod, wherein the members of the ensemble are lazy learning classifierslearnt using the Citation Nearest Neighbour method. Diversity among theensemble members is achieved by optimizing their parameters using amulti-objective optimization method, with the objectives being to maximizeClass 1 accuracy and minimize false positive rate. The method has been found tobe effective on the Musk1 benchmark dataset.
展开▼
