Harnessing data diversity: bridging Kaggle and OpenML repositories

• Grant number: 23/11800-0
• Support Opportunities: Scholarships abroad - Research Internship - Scientific Initiation
• Start date: December 01, 2023
• End date: February 29, 2024
• Field of knowledge: Physical Sciences and Mathematics - Computer Science
• Principal Investigator: Ana Carolina Lorena
• Grantee: André Andrade Gonçalves
• Supervisor: Joaquin Vanschoren
• Host Institution: Divisão de Ciência da Computação (IEC). Instituto Tecnológico de Aeronáutica (ITA). Ministério da Defesa (Brasil). São José dos Campos , SP, Brazil
• Institution abroad: Eindhoven University of Technology (TU/e), Netherlands
• Associated to the scholarship: 23/04911-0 - Gathering meta-data from competitions, BP.IC

Abstract

Machine Learning (ML), a pivotal subset of artificial intelligence, has unequivocally revolutionized computer science research and enhanced immense improvements in technological applications by enabling computers to learn from data, make decisions, and identify patterns autonomously. While various ML algorithms exist, studies have shown and acknowledged that no single algorithm universally excels in every other possible dataset, such that an efficient algorithm selection process is crucial, which can be supported by meta-learning studies. These studies can be used to identify and isolate the main properties of given datasets, mainly for classification and regression tasks, and to incorporate information about ML algorithms performance, a process that requires extensive data and effort to extract the wanted features. Fortunately, public online repositories, such as OpenML and Kaggle, and pyMFE, a tool for data analysis and meta-feature extraction, offer those requirements and can be used together, in complement, to advance meta-learning and AutoML capabilities to confront and overcome their individual limitations. By combining the efforts of the research group responsible for OpenML, represented by Professor Joaquin Vanschoren, and our team related to the FAPESP scientific initiation program, this BEPE proposal intends to enhance meta-learning studies and empower the dataset-gathering process, as well as its characterization and selection. (AU)