Accurate mapping of mechanical properties across extensive areas in heterogeneous materials is essential for understanding phase-specific contributions to strength and hardness. High-speed nanoindentation mapping enables their x-y spatial mapping through a fast and dense grid of indents. However, accurate measurements are complicated by pile-up, the plastic displacement of material laterally and vertically around an indent, causing hardness and modulus overestimation, especially in materials with varying phase compliance. Traditional correction methods rely on time-consuming, localized Atomic Force Microscopy measurements, which are impractical for large-area mapping. This study presents a fast and semi-automated solution using High-speed nanoindentation mapping-induced surface roughness changes Sa, quantifiable by optical profilometry, with machine learning to correct pile-up over extensive areas selectively. By correlating these roughness changes with the Atomic Force Microscopy-measured pile-up height, we derived universal calibration functions for a wide range of bulk materials and thin films, validated through Finite Element Modeling. Applied to a benchmark cobalt-based, chromium-tungsten alloy, the method uses unsupervised clustering to identify piling-up phases in the cobalt matrix while excluding the hard carbides. This approach reduced the hardness and modulus errors by up to 7 %, uniquely enabling accurate phase-specific property mapping in high-speed nanoindentation, advancing the mechanical microscopy frontier.