Forecasting the development of poultry farming based on time series

Purpose. The purpose of this work is to forecast the dynamics of the development of the poultry population for a period of 2 years with the help of various models, which are applied to study time series. Methodology / approach. To conduct a comprehensive study on forecasting the number of poultry population, three predictive models were proposed: two based on regression methods, including SARIMAX and FbProphet, and one with a probabilistic approach using GluonTS. These models were selected to explore different methodological perspectives, ensuring a robust analysis of forecasting accuracy and applicability across varying data patterns and time horizons. To assess the quality of the forecast, the indicators of the mean absolute error, the standard deviation, the mean absolute error in percentage and the mean absolute scaled error for 24 months of forecasting are determined and analysed. The study was conducted based on regional data (using the example of the Khmelnytskyi region of Ukraine). Results. The study successfully applied advanced data science methods to predict changes in poultry population using a number of efficient models. Analysis of historical data allowed us to determine the optimal parameters of the models and obtain forecast values for time periods (months). The studied series of dynamics of monthly changes in the poultry population was tested for stationarity using the Box-Cox transformation. The constructed time series are compared with the actual values, which is illustrated in the graphs. The results demonstrate that the SARIMAX(3,1,2)(1,1,1,12) model provides the best forecast accuracy compared to the other two models, confirming its effectiveness for forecasting tasks. These results highlight the potential of modern forecasting methods in the agricultural sector, offering a data-driven foundation for more effective decision-making in poultry management. Originality / scientific novelty. This study fills a gap in applying advanced forecasting methods to poultry population prediction by systematically comparing SARIMAX, FbProphet, and GluonTS models. Unlike previous research, which often relied on simpler statistical approaches, this study integrates machine learning techniques to enhance forecasting accuracy. The findings confirm an increasing trend in the time series and demonstrate that the SARIMAX model outperforms the alternatives, providing the most precise forecasts for the next two years. Practical value / implications. This study allows poultry farms and enterprises to predict the dynamics of poultry population, which is a critical case for optimising production processes. The use of more accurate forecasting models helps to more effectively plan resources (feed, housing area, personnel), regulate production volumes (eggs, meat), and also ensures supply stability. In addition, the ability to pre-estimate future changes allows enterprises to adapt to market fluctuations, reduce losses, minimise excess costs and make informed management decisions.