The prediction of soil bearing capacity and settlement forms a fundamental aspect of geotechnical engineering, referring to the soil’s ability to withstand applied loads without failure, as well as to the extent of structural settlement or vertical displacement due to soil compression. In this study, considering the complexities involved in soil behavior—particularly under oil contamination—laboratory and field tests in conjunction with machine learning techniques have been employed to improve prediction accuracy. The proposed machine learning approach is based on soft computing methods and incorporates the Flow Direction algorithm as a metaheuristic for parameter optimization in Artificial Neural Networks (ANN) and Support Vector Machines (SVM), proceeding in two main phases. The first phase involves the prediction of internal friction angle and soil cohesion based on influential factors, including oil contamination. The second phase focuses on predicting the bearing capacity as a function of the internal friction angle and cohesion. Additionally, to further enhance prediction accuracy, a multimodal learning technique comprising three Prediction model was utilized, which includes two methods—ANN and SVM—optimized with the improved Flow Direction algorithm and a Decision Tree approach. The results demonstrated that the proposed model could predict bearing capacity with a minimum error of 0.2072, outperforming traditional empirical formulas such as Hansen, Meyerhof, and Vesic.