Consider I am using multiple KDE bandwidths for my data and I would like to find out what the 'best-fit-model' would be. Is there anything like AIC for glm that I could use to accomplish this? I have used AUC and complexity analysis outlined in papers like Silva et al. 2018, but my Reviewer 2 has notified me that AUC isn't sufficient for this. After running the analysis on several nonsensical smoothing values I tend to agree.
Any assistance is appreciated.