A study on the recommendation method to support target variable selection in machine learning

As machine learning is accepted as one of the most powerful tools to solve problems that involve a large amount of data, more and more application areas started to explore and adopt machine learning methods into their workflows. The first step of developing a machine learning model is to determine the target variable, i.e. what to predict. For instance, if a practitioner wants to develop a model to assist hiring decisions, he/she may choose a candidate's performance score as the prediction target to train the machine learning model. In practice, practitioners need to experiment and evaluate a large set of potential target variables until they reach a satisfactory one. We would like to automate the evaluation of target variables and recommend good ones to the users. The purpose of this study is to understand how different recommendation strategies influence the target selection process.