Precision Learner in Classification: A Subject-Based Approach for Classification Using Item Response Theory for Ensemble Machine Learning

• Activity Number: 288 - SLDS CSpeed 5
• Type: Contributed
• Date/Time: Wednesday, August 11, 2021 : 1:30 PM to 3:20 PM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Di Xiong* and Honghu Liu
• Companies: University of California, Los Angeles and University of California, Los Angeles
• Keywords: Classification; Patient-reported Survey; Ensemble Method; Item Response Theory; Machine Learning; Super Learner

Abstract

Machine learning is emerging as an approach to tackle prediction and classification issues in health and biomedical research. It has been proved that the ensembled method can improve machine learning results by combining multiple weaker learners. Most algorithms were designed to optimize overall performance while ignoring the uniqueness of each subject (i.e., patient) and/or classification ability of learners. However, instances can be limited by their responding abilities, especially for patient-reported survey research. Precision Learner aims to tailor machine learning by appropriate subjects’ responding ability and learners’ classification ability using item response theory. An optimal stratification is identified to group subjects by classification performance. Instead of tuning the hyperparameters for each learner, we weight different variants based on their classification abilities in each stratum identified. The proposed algorithm is compared with other widely used methods. This work will provide a comprehensive solution to improves the interpretability of machine learning through balancing prediction accuracy and data complexity.