Shallow-founded buildings are susceptible to liquefaction-induced settlement (Sl) in the event of an earthquake. Mitigating earthquake damage requires accurate settlement evaluation. Nnonetheless, the process of predicting the Sl is not simple and necessitates advanced soil models and calibrated soil characteristics, which are not easily accessible for specialists and designers. Furthermore, multivariate adaptive regression splines or conventional regression analysis were used to build the available empirical models to estimate the Sl, and these methods result in complex models. Moreover, these empirical models were created by applying the outcomes of numerical modelling. In order to overcome these constraints, this research presents the development of two novel decision tree models: the reduced error pruning (REP) tree, the random forest (RF), and the random tree (RT). The Sl may be immediately and accurately estimated with the new models, which have been developed using authentic laboratory observations from centrifuge results. The data utilized in this research includes seven characteristics: the width of the foundation, the height of the building, the pressure exerted on the foundation, the thickness and relative density of the liquefiable layer, and the intensity of the earthquake. Two subsets of the available data are used: the training set (20%) and the test set (80%). Statistical measures such as root mean squared error, mean absolute error, and coefficient of correlation are utilized to assess the decision tree models' output. Applications of the previously outlined method for predicting the Sl are compared and discussed. The evaluation of the Sl dataset's statistical metrics indicates that the RT produced significantly more dependable and reliable outcomes.