Multiple Instances Learning Using K-Nearest Neighbor

Abstract

Multi-instance learning is a variant of supervised machine learning where learner receives set of bags associated with binary label   rather than each instance is labeled positive or negative. In this learning, each sample bag may has alternate feature vector that depicts it and still only one of those may be responsible for observed label of bag. So a sample is labeled as a positive bag if at least one of its instances is positive. Otherwise, it is labeled as a negative bag. MIL getting growing attention because if its suitability in numerous real world tasks such as image classification, molecular activity prediction, text or document categorization etc. In this paper, the problem definition, learning algorithm and experimental data sets related to multi-instance learning framework are briefly reviewed. The purpose of this work is to show that kNN based MIL may improved results of classification as compared other algorithms. To validate the approach Musk dataset is taken as a benchmark dataset.