A Study on Mitigating Hard Boundaries of Decision-Tree-based Uncertainty Estimates for AI Models