Eco-labelling of food products is critical in promoting sustainable consumption. However, traditional methods of assessing environmental impact for such eco-labelling rely on comprehensive Life-Cycle Assessment (LCA) data, which is costly and resource-intensive to gather. To address this gap, we have leveraged ingredient lists - an underexplored yet widely available data source - to predict the environmental footprint of food products. While previous research has focused predominantly on using product names for this purpose - with only one known study in this domain - our work is one of few to explore the use of ingredient data. Using the Open Food Facts database and Machine Learning (ML) models, including K-Nearest Neighbors (KNN), Random Forest (RF), Ridge Ordinal Regression, Support Vector Regression (SVR) Ordinal, Cumulative Ordinal Regression for Logistic models (CORAL), CORN Ordinal Logistic Regression (CORN), and Extreme Gradient Boosting (XGBoost) variants, we demonstrate that ingredient-based models can effectively predict eco-grade labels, and are particularly strong at distinguishing products in the highest and lowest categories. Our findings highlight the potential of ingredient lists as a reliable proxy that, when combined with other product information, could have high prediction accuracy. This approach provides a scalable approach for the sustainability assessment of food products, offering practical benefits for consumers, policymakers, and industry stakeholders.