Publicação
Automatic Handling of Imbalanced Datasets for Classification
| Resumo: | Imbalanced data is present in various business areas and when facing it without proper knowledge, it can have undesired negative consequences. In addition, the most common evaluation metrics in machine learning to measure the desired solution can be inappropriate and misleading. Multiple combinations of methods are proposed to handle imbalanced data however, often, they required specialised knowledge to be used correctly. For imbalanced classification, the desire to correctly classify the underrepresented class tends to be more important than the overrepresented class, while being more challenging and time-consuming. Several approaches, ranging from more accessible and more advanced in the domains of data resampling and cost-sensitive techniques, will be considered to handle imbalanced data. The application developed delivers recommendations of the most suited combinations of techniques for the specific dataset imported, by extracting and comparing meta-features values recorded in a knowledge base. It facilitates effortless classification and automates part of the machine learning pipeline with comparable or better results to a state-of-the-art solution and with a much smaller execution time |
|---|---|
| Autores principais: | Vieira, Pedro Marques |
| Assunto: | Imbalanced Classification Handling Imbalanced Data Automated Machine Learning Classificação Não Balanceada Manipulação de Dados Não Balanceados Automatização de Aprendizagem de Máquina |
| Ano: | 2022 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Instituto Politécnico do Porto |
| Idioma: | inglês |
| Origem: | Repositório Científico do Instituto Politécnico do Porto |
| Resumo: | Imbalanced data is present in various business areas and when facing it without proper knowledge, it can have undesired negative consequences. In addition, the most common evaluation metrics in machine learning to measure the desired solution can be inappropriate and misleading. Multiple combinations of methods are proposed to handle imbalanced data however, often, they required specialised knowledge to be used correctly. For imbalanced classification, the desire to correctly classify the underrepresented class tends to be more important than the overrepresented class, while being more challenging and time-consuming. Several approaches, ranging from more accessible and more advanced in the domains of data resampling and cost-sensitive techniques, will be considered to handle imbalanced data. The application developed delivers recommendations of the most suited combinations of techniques for the specific dataset imported, by extracting and comparing meta-features values recorded in a knowledge base. It facilitates effortless classification and automates part of the machine learning pipeline with comparable or better results to a state-of-the-art solution and with a much smaller execution time |
|---|