Forecasting Industrial Production: A Comparative Study Based on Volatility and Seasonality of Time Series

Accurate forecasting of industrial production is a key factor for effective planning of production processes, resource allocation, and inventory management. This study uses a curated time series dataset of historical production data from multiple industrial sectors to evaluate the performance of various forecasting approaches. Several statistical and machine learning methods are systematically compared, emphasizing how time series characteristics such as volatility and seasonality influence predictive accuracy across short-, medium-, and long-term horizons. To enhance performance, an ensemble model was constructed by combining selected methods. The results highlight that specific approaches, notably XGBoost, ARIMA, and the ensemble model, achieve consistently high accuracy on stable and strongly seasonal series, while other methods underperform. Furthermore, the analysis reveals that medium-volatility series can yield unexpectedly accurate long-term forecasts, underscoring the importance of considering time series structure in industrial production modeling. This study contributes to modeling and simulation literature by demonstrating how data characteristics shape the effectiveness of forecasting methods in industrial contexts.