A benchmark and comparison of active learning for logistic regression. (November 2018)

• Record Type: Journal Article
• Authors: Yang, Yazhou; Loog, Marco
• Is Part Of: Pattern recognition. Volume 83(2018:Nov.)
• Page Start: 401
• Page End: 415
• Publication Date: 2018-11
• Subjects: Active learning -- Logistic regression -- Experimental design -- Benchmark -- Preference maps; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2018.06.004
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: A review of the state-of-the-art active learning algorithms built on logistic regression is presented, in which links and relationships between methods are explicated. A preference map is proposed to reveal characteristic similarities and differences of the selection locations in 2D problems. Extensive experiments on 44 real-world datasets and three artificial sets are carried out. Insight is provided for the behaviors of classification performance and computational cost. Abstract: Logistic regression is by far the most widely used classifier in real-world applications. In this paper, we benchmark the state-of-the-art active learning methods for logistic regression and discuss and illustrate their underlying characteristics. Experiments are carried out on three synthetic datasets and 44 real-world datasets, providing insight into the behaviors of these active learning methods with respect to the area of the learning curve (which plots classification accuracy as a function of the number of queried examples) and their computational costs. Surprisingly, one of the earliest and simplest suggested active learning methods, i.e., uncertainty sampling, performs exceptionally well overall. Another remarkable finding is that random sampling, which is the rudimentary baseline to improve upon, is not overwhelmed by individual active learning techniques in many cases.