Effectiveness of Automatic Data Transformation in Deep Learning Model for Leasing Decision Support Process

Databases used in machine learning for management decision support require, first of all, adequate preparation: transformation. This process involves determining the type of data, excluding missing data, formatting the data and preparing the text data for recoding. This stage is essential, as it significantly impacts the performance of machine learning. It is usually carried out manually or “semi-automatically”, which is often time-consuming and requires advanced statistical and expert knowledge. The purpose of this article is to describe the method of automatic data transformation and to show that this method gives as good results as manual transformation, however, it allows to perform the operation without unnecessary waste of time and effort. The analysis was carried out using a dataset obtained from leasing companies, it contains a wide variety of data, many missing data and abounds in unique or categorizing text data. The results of applying the automatic transformation algorithm are based on a previously designed simulation experiment. Measurements of machine effects were made based on measures of the effectiveness of machine learning. The measurements show that the automatic transformation produces statistically indistinguishable machine-learning effects as the manual transformation.