The use of deep cement mixing (DCM) columns is an effective and affordable technique for ground stabilization. However, designing this method can be complex due to uncertainties in the geotechnical properties of the soil and DCM columns, area improvement ratio, column arrangement, and required cement content. This study aimed to address this issue by using Gaussian process regression (GPR) models to estimate the ultimate bearing capacity (UBC) of soft soil improved with DCM columns.To create and train the GPR models, the study utilized a database of 46 physical modeling tests under end-bearing and floating conditions. The researchers used different kernel functions, including rational quadratic, squared exponential, Matern 5/2, and exponential, for the GPR models. The models were then optimized through Bayesian optimization and compared to other predictive techniques such as multilayer perceptron (MLP), radial basis function (RBF), and neuro-fuzzy inference systems (ANFIS) using test data.As a case study, the researchers evaluated a decision-making model for designing the geotechnical properties of DCM columns. The results showed that the optimized GPR model's accuracy in terms of performance indices was satisfactory for both end-bearing and floating DCM column conditions. The optimized GPR model outperformed MLP, RBF, and ANFIS performance indices using test data. Overall, the study demonstrated that optimized GPR models are a promising method for early prediction of stabilized ground UBC.