Abstract: Detecting pile foundation depths is essential for engineering construction in urban areas. However, traditional approaches for this detection mostly exhibit insufficient accuracy. Pile foundations are primarily composed of concrete and steel bars, displaying distinct electrical conductivity from the background values. Electrical methods are sensitive to subsurface conductivity variations, thus providing a novel approach for detecting pile foundation depths. Focusing on typical pile foundations used in the engineering construction of Guangzhou Metro, this study evaluated the effectiveness of dipole-dipole and Wenner configurations in detecting pile foundation depths through numerical simulations. Furthermore, the pile foundation depths were predicted using a random forest model. The results indicate that both configurations could detect the potential differences and apparent resistivity responses of pile foundations at burial depths not exceeding 30 m. After training with 1000 samples, the random forest model accurately and rapidly predicted the pile foundation depths while showing strong robustness in resistance to noise. Overall, the borehole electrical method exhibits significant potential for detecting pile foundation depths, which, combined with the random forest model, offers an efficient tool for accurate prediction of pile foundation depths.