Abstract
The article explores the features of applying gradient boosting algorithms based on decision trees for time series sales forecasting in the automotive parts retail sector. Approaches to the development of a system for gross revenue forecasting using the ML.NET platform and the FastTree Regression model are considered. The aim of the study is to develop an information and analytical system for sales time series forecasting using the FastTree Regression gradient boosting algorithm to account for seasonal fluctuations and irregular demand in the automotive parts market. The proposed approach is aimed at improving the efficiency of transactional data analysis and supporting financial planning processes within enterprises. The obtained results can be used to automate sales forecasting and enhance business analytics systems. Methods for preprocessing transactional data, anomaly detection, and generating time-based features are analyzed. The paper proposes an improved approach to building and normalizing seasonal features within the data processing pipeline, which increases the model’s adaptability to irregular demand. Unlike standard approaches limited to the use of static time-related features, the proposed solution provides automated generation of seasonal characteristics during the forecasting process. To ensure system performance efficiency, the analytical modules are integrated directly with data filtering, processing, and aggregation mechanisms at the database level. Mechanisms for primary data import, saving results in JSON format, and automated generation of forecasted values for a given planning horizon are implemented. The model was validated using historical sales data for the 2023–2026 period, and the forecasting quality was evaluated using statistical metrics. The obtained results confirm the applicability of the developed approach for automated analysis of sales dynamics and forecasting of future revenue indicators in retail environments. The obtained value of the coefficient of determination (R² = 0.90) and RMSE = 287,883 UAH demonstrate the effectiveness of ensemble methods for time series sales forecasting under conditions of high demand variability. The developed web system based on the ASP.NET Core MVC architecture and interactive visualization tools can be effectively used in practical processes of financial analysis, procurement volume planning, and management decision-making of retail companies.