Particle Swarm Optimization Based Swarm Intelligence for Active Learning Improvement: Application on Medical Data Classification

摘要: Semi-supervised learning targets the common situation where labeled data are scarce but unlabeled abundant. It uses to help supervised tasks. In practice, it may make sense utilize active in conjunction with semi-supervised learning. That is, we might allow algorithm pick a set of instances be by domain expert, which will then used as set. However, existing approaches computationally expensive and require searching through an entire dataset, contain redundant that provide no instructive information classifier can decrease performance. To address this optimization problem, hybrid system combines (AL) particle swarm (PSO) algorithms is proposed reduce cost labeling while building more efficient classifier. The novelty work resides integration bio-inspired machine strategy. Furthermore, novel uncertainty measure was integrated into objective function select from massive amounts medical those deemed most informative. evaluate effectiveness approach, eighteen (18) benchmark datasets were compared against three best-known classifiers different paradigms: AL–NB using Naive Base Margin Sampling strategy, SVM (Support Vector Machine), ELM (Extreme Learning Machine) learning, TSVM (Transductive Support Experiments showed approach effective reducing efforts required experts for annotation produce accurate has been utilized optimize task labeling. Based on measure, nature-inspired PSO attempts considered informative, at same time improving experiments carried out confirm strategy significantly enhances performance AL commonly strategies. achieves similar fully requiring much less As future extension work, would interesting integrate other evolutionary compare them our approach. addition, beneficial test impact variants Also, aimed classification experimentation process.