Instance level hardness analysis for multilabel problems

• Grant number: 25/10035-3
• Support Opportunities: Scholarships in Brazil - Master
• Start date: September 01, 2025
• End date: August 31, 2027
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Ricardo Cerri
• Grantee: Bruno Uhlmann Marcato
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
• Associated research grant: 22/02981-8 - Novelty detection in multi-label data streams classification, AP.PNGP.PI

Abstract

Data classification is one of the main tasks in Machine Learning (ML), consisting of assigning unknown patterns to one or more known classes. In multilabel problems, where an instance can simultaneously belong to multiple classes, this task poses additional challenges that make its modeling more complex. Several algorithms have been applied to multilabel classification, either as adaptations of existing methods or as algorithms that transform the problem into simpler tasks. Likewise, specific metrics are required to properly evaluate the models generated. In ML experiments, evaluating only global metrics can obscure systematic errors in subsets of the data, making a more refined analysis necessary. In this context, measures that quantify the difficulty of correctly classifying instances can be used to improve the understanding of classifier behavior and guide optimization strategies. To address the challenges of multilabel classification, this project proposes the development and investigation of new instance level hardness metrics specifically tailored for this type of problem. The main objective is to investigate the concept of instance hardness in multilabel classification problems and to propose ways to measure it. (AU)