Development of an Automatic learning and Validation Algorithm for White-box Binary Classification Models

The main purpose of the AutoML systems’ development is to let the person who is not specialist in ML to create ML models, without having a wide range of competencies in the field of machine learning. Currently, there are many implementations of AutoML algorithms. All of them are designed to maximize the quality of model predictions, so they include complex elements. These include intensive hyperparameter optimization, the use of black box algorithms, stacking, etc. However, for some fields (such as banking or medicine) it is important to obtain so-called interpreted models, the mechanism of which is relatively transparent to the user. Such models include, for example, models based on the logistic regression algorithm or a shallow decision tree. This paper focuses on the creation of learning and validation algorithm of white-box models in order to obtain high-quality, but interpreted results with the support of any quality metrics and validation schemes. The results will be used in the validation of models in the banking business.