arXiv:2405.15950v1 Announce Sort: new
Summary: Machine studying fashions for steady outcomes usually yield systematically biased predictions, significantly for values that largely deviate from the imply. Particularly, predictions for large-valued outcomes are usually negatively biased, whereas these for small-valued outcomes are positively biased. We check with this linear central tendency warped bias because the “systematic bias of machine studying regression”. On this paper, we first display that this difficulty persists throughout numerous machine studying fashions, after which delve into its theoretical underpinnings. We suggest a normal constrained optimization method designed to right this bias and develop a computationally environment friendly algorithm to implement our technique. Our simulation outcomes point out that our correction technique successfully eliminates the bias from the anticipated outcomes. We apply the proposed method to the prediction of mind age utilizing neuroimaging information. Compared to competing machine studying fashions, our technique successfully addresses the longstanding difficulty of “systematic bias of machine studying regression” in neuroimaging-based mind age calculation, yielding unbiased predictions of mind age.
Supply hyperlink