To help earthquake early warning (EEW) systems make quick decisions, we build a random forest (RF) model for rapid earthquake localization. This system computes the differences in P-wave arrival timings between the first five stations to record an earthquake as a reference station (i.e., the first recording station). The RF model categorises these differential P-wave arrival times and station locations in order to determine the epicentral position. Using a Japanese earthquake catalogue, we train and evaluate the suggested algorithm. The Mean Absolute Error (MAE) of the RF model, which forecasts earthquake sites, is 2.88 km. Importantly, the suggested RF model can learn from little data— 10% of the dataset—and a lot fewer recording stations—three—and yet get good results (MAE5 km). The approach provides a potent new tool for quick and precise source-location prediction in EEW since it is accurate, generalizable, and responsive