A carregar...
Publicação
A Structured Framework for AutoML: Integrating LLMs through Comparative Experiments
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 924.09 KB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
This work explores the potential constructive interaction between Generative AI, specifically
ChatGPT-4, and Automated Machine Learning (AutoML) frameworks. The study focuses on
leveraging ChatGPT-4's capabilities within the CRISP-DM (Cross-Industry Standard Process for Data
Mining) model phases to improve the efficiency and effectiveness of data-driven tasks. Through a
series of experiments involving classification, regression and clustering, the research compares the
performance of ChatGPT-4 in two settings: a global user perspective with general prompts and a
structured approach aligned with the CRISP-DM methodology, providing a comparative benchmark.
The findings demonstrate that aligning ChatGPT-4's tasks with the CRISP-DM phases yields better
performance and more comprehensive insights than the general prompt approach. The study
highlights the importance of prompt engineering in optimizing ChatGPT-4's contributions to AutoML
tasks, emphasizing its role in improving data preparation, model selection and the evaluation
processes. Additionally, the research underscores the ethical considerations and potential challenges
associated with integrating generative AI in AutoML, particularly concerning data quality, bias, and
model interpretability.
Descrição
Dissertation presented as the partial requirement for obtaining a Master's degree in Information Management, specialization in Knowledge Management and Business Intelligence
Palavras-chave
Generative AI AutoML Prompt engineering ChatGPT Machine learning techniques CRISP-DM SDG 8 - Decent work and economic growth