How does Random Forest behave in high-dimensional low-sample-size (HDLSS) regimes?

Random Forest may overfit in HDLSS settings due to sparse sampling of feature space.

In HDLSS problems (p >> n), each tree may find spurious splits due to high dimensionality. Although feature subsampling reduces risk, instability remains because training data is insufficient to represent true distributions. Feature importance becomes highly unstable, and variance across trees increases.