Before anything else, I would start with the basics and try to make something that can reliably describe the overall sound signature of a headphone based on its graph. As is, the execution seems wrong on many levels.

You should not be examining portions of the frequency range solely in isolation, as bass/midrange/treble exist relative to each other and each affect how the others sound, you should be aware of how the measurements are being compensated, you should know the limitations and quirks of various measurement setups, you should 100% not be taking MiniDSP EARS treble measurements at face value regardless of how it’s calibrated. Even if the AI were not hallucinating a peak at 10 kHz, “a peak at 7 kHz followed by a sharp dip and a peak at 10 kHz” describes what the pinna notch might look like on a typical GRAS rig. As questionable as AutoEQ may be for actually improving sound, its measurement database would be a much better means of sourcing standardized graphs than arbitrary user-uploaded images.

You need to use your listening impressions of the headphones whose FRs you’re running through this program as a sanity check to ensure it isn’t spouting nonsense. There is no getting around making the effort to correlating FR features to what you are hearing in reality if you don’t want the output to be a crapshoot. Assuming you wrote the Headphonesty review this graph originates from, you yourself are calling the H1H’s midrange mediocre at best and the treble subdued.