How does Random Forest relate to functional estimation in L2 function spaces?

Random Forest can be interpreted as a nonparametric estimator converging in L2 space to the regression function.

In L2(R^d), Random Forest acts as an estimator f̂(x) that minimizes expected squared error E[(Y - f̂(X))²]. Each tree defines a step function over partitions of feature space. The ensemble averages these functions, producing a projection-like operator in L2 space. Under certain regularity conditions, RF is consistent, meaning f̂ → f = E[Y|X] in L2 norm as number of samples and trees grow.