Unlocking The Potential Of Machine Learning In The Derivation Of Low-Energy Models For Metallic Magnets