<p>The significance of this work lies in its innovative approach to improving the accuracy of predictive modeling in complex systems, particularly in estimating the settlement (S_m), a parameter critical to the research domain. By integration models, the study addresses the limitations of traditional models that rely on predefined formulations, enabling more precise modeling of nonlinear relationships between input and output variables. This approach not only enhances the reliability of predictions but also broadens our understanding of the underlying dynamics, making it highly valuable for applications that require robust and accurate forecasting. The objective of this study is to implement recently developed machine learning models, namely hybridized support vector regression (SVR) and multi-layered perceptron (MLP) with artificial rabbit optimization (ARO), as effective methods for forecasting the S_m of shallow foundations based on cohesion soil properties. Optimization techniques facilitated the determination of the optimal value for the primary variable in the SVR model and the optimal number of neurons for each hidden layer in the multi-layer perceptron MLP. The study’s findings indicate that the ARMLP and ARSVR systems, when used in combination, exhibit adept estimation capabilities. This is supported by the R² values obtained from the ARSVR’s training and testing components, which were 0.9856 and 0.9793, in that order, and the ARMLP’s R² values of 0.9732 and 0.9719 for its testing and training components. Regarding the ARSVR model, NRMSE metrics reduced from 0.224 to 0.1633 in the training phase and from 0.213 to 0.184 in the test stage. Ultimately, the work significantly advances the state of the art, offering new insights and tools that can be applied across various disciplines where predictive accuracy is essential.</p>