Comparison of Ensemble-Based Multiple Instance Learning Approaches

Abstract

Multiple instance learning (MIL) is concerned with learning from training set of bags including multiple feature vectors. This paradigm has various algorithms as a solution for multiple instance problem. Recently, ensemble learning has become one of the most preferred machine learning technique because its high classification ability. The main goal of ensemble learning is combining multiple learning models and obtaining a decision from all outputs of these models. Considering this motivation, the study presented in this paper proposes an ensemble-based multiple instance learning approach which merges standard algorithms (MIWrapper and SimpleMI) with ensemble learning methods (Bagging and AdaBoost) to improve classification ability. The proposed approach includes ensemble of combination of MIWrapper and SimpleMI learners with Naive Bayes, Support Vector Machines (SVM), Neural Networks (Multilayer Perceptron (MLP)), and Decision Tree (C4.5) as base classifiers. In the experimental studies, the proposed ensemble-based approach was compared with individual MIWrapper and SimpleMI algorithms in terms of accuracy. The obtained results indicate that the ensemble-based approach shows higher classification ability than the conventional solutions.