CART-ROSA Framework Opens Random Forest Black Box with Stochastic Control Theory

A new paper by Tianxing Mei, Yingying Fan, Mingming Leng, and Jinchi Lv (arXiv:2605.26675) proposes CART-ROSA (CART Random Opportunity-Set Allocation), a stochastic-control framework that demystifies the internal mechanics of CART random forests. At each node, the random subset of features is treated as a random feasible action set, and the CART split rule as a masked-action allocation policy. This induces a controlled stochastic process over informative split-count states, whose terminal law determines both single-tree error and cross-tree interaction terms in the forest mean squared error (MSE). The framework separates two design levers: the informative-opportunity rate (driven by feature subsampling) and the contraction strength (from the within-mask split policy).

The authors establish that the CART policy is locally stabilizing—it contracts imbalances in informative split allocations and concentrates terminal tree geometry. However, at the system level, it can be globally suboptimal for the forest objective. Specializing to a linear model, they derive the MSE risk expansion explicitly. This operations-research perspective makes tractable a theoretical gap that was previously difficult to access from standard algorithmic descriptions. The 69-page paper includes one figure and provides a foundation for more principled tuning of random forest hyperparameters, such as mtry and tree depth, based on a clear theoretical understanding of ensemble risk.