A review on preprocessing algorithm selection with meta-learning

Several AutoML tools aim to facilitate the usability of machine learning algorithms, automatically recommending algorithms using techniques such as meta-learning, grid search, and genetic programming. However, the preprocessing step is usually not well handled by those tools. Thus, in this work, we present a systematic review of preprocessing algorithms selection with meta-learning, aiming to find the state of the art in this field. To perform this task, we acquired 450 references, of which we selected 37 to be evaluated and analyzed according to a set of questions earlier defined. Thus, we managed to identify information such as what was published on the subject; the topics more often presented in those works; the most frequently recommended preprocessing algorithms; the most used features selected to extract information for the meta-learning; the machine learning algorithms employed as meta-learners and base-learners in those works; and the performance metrics that are chosen as the target of the applications.