Random Forest multiclasse: a diagnostic study of mathematical learning difficulties

• Ano de defesa: 2024
• Autor(a) principal: Augusto, Patrícia Bruniero Franciscato
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Diagnosis; Diagnóstico; Dificuldade de aprendizagem; Discalculia; Dyscalculia; Learning disability; Matemática; Mathematics; Psicometria; Psychometrics; Random Forest
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/59/59143/tde-08032024-073933/

Specific learning disorders (SLD) have a neurobiological origin and are classified according to their specific domains. Developmental dyscalculia (DD) is a SLD with persistent academic impairments in mathematical skills regarding numerical sense, memorization of arithmetic facts, performance or fluency of calculations and mathematical reasoning. The development of efficient diagnostic mechanisms for DD using machine learning techniques has gained significant attention in recent research. Conventionally, the diagnosis of DD involves time-consuming processes, including multiple tests and interviews that extend over weeks or months. However, recent studies have demonstrated the potential for generating classifier models with high performances using psychometric instruments, which can contribute to reducing the complexity of the diagnostic process. This research presents a framework to identify opportunities to the NUMERO Outpatient Clinic protocol using Random Forest for classification and variable ranking analyses. Applying a dimensionality reduction mechanism, a hybrid method combining hierarchical clustering and RF classification, we proposed to eliminate irrelevant variables and, consequently, largely improve model\'s efficiency. Computer simulations present promising results throughout many dataset versions. Our approach holds great potential for efficiently support diagnosing developmental dyscalculia, offering a valuable contribution to the field of cognitive assessment and intervention, while may also be adapted to another psychometric based diagnose.