Avaliação de técnicas de geração de dados sintéticos tabulares em tarefas de classificação

Abstract

The generation of synthetic data plays a crucial role in machine learning across various domains. This study explores SDV, a tool that utilizes machine learning methods to produce high-quality synthetic data. To achieve this, a complete data pipeline is developed to ensure the automated and accurate persistence of training and testing data, covering the generation, application, and evaluation of synthetic data in ML algorithms for classification. The effectiveness of these steps is verified through comparisons between different strategies for creating synthetic data generator models and their use in classification models, using metrics suitable for each approach. By employing SDV on different datasets, this research highlights the relevance and utility of creating and using artificial data, enabling more robust analyses and more accurate machine learning models. Thus, SDV provides a promising solution to address the challenges of data scarcity, imbalance, and privacy, driving research and applications in the field.