This paper develops a machine learning framework for detecting and predicting liquidity sweep events in XAUUSD using event-based market microstructure analysis. Using 15-minute data from 2014–2024, the study formalizes liquidity sweeps as a binary classification problem evaluated through walk-forward validation. A calibrated Random Forest classifier demonstrates statistically meaningful predictive structure while also revealing a significant prediction–execution gap: predictive accuracy does not necessarily translate into profitable trading performance under naive execution. The work contributes to quantitative finance, market microstructure research, and machine learning-based financial forecasting.
Vanshvardhan Sharma (Sat,) studied this question.