Multiple Instance Learning Based on Twin Support Vector Machine

摘要

Each input object in multiple instance learning (MIL) is represented by a set of instances, referred to as 'bag.' Therefore, in MIL, class labels are associated with each bag instead of individual instance. This study proposes a classifier for multiple instance learning based on Twin Support Vector Machine, termed as MIL-TWSVM. The proposed approach is trained at bag level, where each bag is represented by a vector of its dissimilarities to other bags in the training set. A comparative analysis of MIL-TWSVM approach is performed with the instance-level and noisy-or (NOR) learning approaches based on TWSVM. The performance of the proposed MIL-TWSVM approach has also been compared with several existing approaches of multiple instance learning. The experiments on eight multiple instance benchmark datasets have shown the superiority of the proposed approach. The significance of experimental results has been tested via statistical analysis conducted by using Friedman's statistic and Nemenyi post hoc tests.