Bearing capacity estimation of shallow foundations is the essential requirement in the design of structures and taking a calculation method into account is necessary. All the parameters and uncertainties cannot be factored in by the classic analytical-based methods. Moreover, performing in-site tests require an extensive period of time and many resources. With the development of new methods such as Machine Learning (ML) algorithms in recent decades, a resolution to these challenges has been identified. In this study, classic machine learning regression methods such as KNN, SVM and Decision Tree based models alongside the utilization of Artificial Neural Networks (ANN) regression are examined and compelling results are demonstrated. The dataset in this study is consisting of 97 tests on model foundations and site loadings on granular soil. The results indicate that ML regression methods will have reliable outcome in determination of bearing capacity. But more importantly, the precision of the trained model is closely correlated to data splitting and the ratio of train and test series in the dataset. The importance of splitting procedure was examined through trial and error with parameters of train test data ratio and the random state of sampling. It is indicated that a ratio of 80% for the training set would be an optimum value. Furthermore, relative importance of the input features was examined through a sensitivity analysis which indicated that the internal friction angle of the soil and the depth of the foundation are the most important inputs while using ML regression methods.