Forecast accuracy is a binding constraint on inventory performance for intermittent demand patterns, which are common in spare parts inventories, according to Syntetos et al.~\cite{ref11}, inspiring techniques that go beyond basic moving averages. Huber and Stuckenschmidt~\cite{ref12} showed that ensemble approaches combining both paradigms consistently outperform either alone, and that Long Short-Term Memory (LSTM) networks capture temporal dependencies in demand time series that tree-based models miss. The current paper fills a research gap regarding the incorporation of sustainability considerations into such machine learning models.

According to Syntetos et al., forecast accuracy creates a significant limitation on inventory performance when dealing with intermittent demand patterns, which frequently occur in spare parts inventories. This challenge has led to the development of forecasting techniques that extend beyond simple moving averages. Research by Huber and Stuckenschmidt demonstrated that ensemble methods that combine both approaches reliably deliver superior results compared to using either method independently. Their work also revealed that Long Short-Term Memory (LSTM) networks effectively identify temporal dependencies within demand time series that tree-based models fail to detect. This paper addresses an existing research gap by exploring how sustainability factors can be integrated into these machine learning models.