BAdaCost: Multi-class Boosting with Costs. (July 2018)

• Record Type: Journal Article
• Authors: Fernández-Baldera, Antonio; Buenaposada, José M.; Baumela, Luis
• Is Part Of: Pattern recognition. Volume 79(2018:Jul.)
• Page Start: 467
• Page End: 479
• Publication Date: 2018-07
• Subjects: Boosting -- Multi-class classification -- Cost-sensitive classification -- Multi-view object detection; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2018.02.022
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: Boosting algorithm for solving multi-class cost-sensitive problems. The cost matrix can be used as a tool to learn class boundaries (i.e class imbalance, detection problems, Jaccard or F1 objective problems, etc). Multi-class object detectors can be developed with BAdaCost resulting in faster performance than the one-vs-background usual approach (using binary detectors). The cost matrix can be used to improve Average Precision (AP) in multi-view object detection problems. Graphical abstract: Abstract: We present BAdaCost, a multi-class cost-sensitive classification algorithm. It combines a set of cost-sensitive multi-class weak learners to obtain a strong classification rule within the Boosting framework. To derive the algorithm we introduce CMEL, a Cost-sensitive Multi-class Exponential Loss that generalizes the losses optimized in various classification algorithms such as AdaBoost, SAMME, Cost-sensitive AdaBoost and PIBoost. Hence unifying them under a common theoretical framework. In the experiments performed we prove that BAdaCost achieves significant gains in performance when compared to previous multi-class cost-sensitive approaches. The advantages of the proposed algorithm in asymmetric multi-class classification are also evaluated in practical multi-view face and car detection problems.