<p>A good knowledge of subsurface conditions is crucial for underground construction and risk management. When boreholes are sparse, geophysical data can be a valuable supplement for stratigraphic delineation. However, geophysical results can be affected by external factors, and a reliance solely on geophysical and borehole data may introduce errors and uncertainties in geological interpretation. This study proposes an approach that utilizes multi-source geodata and the interpretation enhanced with machine learning (ML) and spatial correlation for geological profiling and uncertainty assessment. In this approach, an iterative interpolation method is first employed to augment the training dataset quantity. The augmented dataset with geophysical and spatial correlation features is then used for geological profiling through stacking ensemble learning. The prediction uncertainty can be assessed using the probability outputs from the ML models involved in stacking ensemble learning. The proposed approach was first validated using a synthetic case and applied subsequently to a site in Singapore. For the synthetic case, the determined geological cross-section achieves a 93.10% accuracy compared to the predefined cross-section. The spatial correspondence between low integrated confidence (IC) values and misclassification areas confirms the IC plot as an effective tool for uncertainty assessment and identification of likely prediction errors. For the site in Singapore, the proposed approach can delineate the boundaries of four geological formations and provide valuable insights into rockhead variations. The discrete cells with low IC values reasonably indicate potential variations of stratigraphic interfaces. Validation using three boreholes shows a prediction accuracy of 87.85%, demonstrating the benefit of multi-source geodata integration in improving geological profiling reliability.</p>