Data analysis and the ability to preserve user privacy has become very important nowadays. So the data is coded without losing important information and made available for research. Gradient-boosted decision trees provide competitive, robust regression and classification procedures that can be interpreted and implemented to predict cold-start processes and are particularly suitable for working with less clean data. The LightGBM method is generally used for classification problems, both for open and encoded data. It is proposed to apply this method for forecasting the demand price of a product without history. We propose to use LightGBM as a safe and fast implementation of a loaded decision tree algorithm, which is widely used in data mining and machine learning tasks. For research, it is important to have a fairly wide base of goods with certain categories and goods with detailed characteristics.