In this study we aim to predict missing numerical attribute values for entities in KGs by incorporating arithmetic relationships between attributes as constraints in the models. The expectation is that these constraints will act as structural regularization to achieve a consistent result.