MLtool: A Tool to Automate the Construction, Evaluation, and Selection of Machine Learning Models

Abstract

The use of Machine Learning has intensified in recent years, gaining notoriety in the most diverse applications. Thus, there is a growing demand for professionals in this area, which has favored the entry of countless inexperienced users in the labor market. The path from the initial step to the use of algorithms and Machine Learning techniques in production environment takes considerable time, even from experts, due to the realization of interactive and manual tasks required for building predictive models that provide the desired results. In view of this, this work presents a tool to automate the construction, evaluation and selection of predictive models considering Machine Learning algorithms and parameters values more appropriate for each situation. The stages of feature selection, standardization, resampling, and training in the construction of models were considered in the tool. The proposed tool deal with the treatment of the problem of data unbalancing, as needed, as well as the execution control of the steps involved in the process of creating predictive models. The results obtained demonstrate that the order and choice of the steps and the values chosen for the algorithm parameters affect the final results of the generated models.