A Systematic Review of Automated Feature Engineering Solutions in Machine Learning Problems

• Fernando Favoretti Vital Prado USP
• Luciano Antonio Digiampietri USP https://orcid.org/0000-0003-4890-1548

DOI: https://doi.org/10.5753/sbsi.2020.13756

Resumo

During the last decades, machine learning has played an important role in building data-driven experiments. Based on information extracted from a variety of sources, new patterns can be identified, predictions can be made more easily and decisions made faster and more effective. The specialized application of machine learning solutions requires specific knowledge in areas such as math, computation, and statistics, as well as being extremely costly in time and having a high chance of incurring any kind of human error during the process. Automated Machine Learning Techniques (AutoML) seek to automate parts of the process of building machine learning applications, allowing non-experts to perform this process. An important part of this kind of problem is the feature engineering part which creates a transformation in the data, making it more representative for the final model. This paper presents a systematic review of automated feature engineering solutions in machine learning problems. With the main objective of identifying and analyzing the existing methods and techniques for performing the automated feature engineering step within the framework of machine learning problems.

Palavras-chave: Auto Learning, Feature Engineering, Feature Selection, Feature Generation