Accurate prediction of soil liquefaction is critical for ensuring geotechnical safety in earthquake-prone regions, and recent advances in machine learning (ML) offer promising alternatives to traditional empirical and reliability-based methods. In this study, four machine learning (ML) classification models including Decision Tree (DT), Naive Bayes (NB), and their respective Principal Component Analysis (PCA)-based hybrid variants, PCA-DT and PCA-NB, are developed to assess soil liquefaction susceptibility. The predicted accuracy of the suggested models is assessed and compared using several performance parameters including Accuracy, F1 Score, Matthews Correlation Coefficient (MCC), Geometric Mean (Gmean), Sensitivity, Specificity, Balanced Accuracy (BA), and AUC. Results show that the accuracy of the DT model was the most accurate with an accuracy of 90% and 84.5% for the training and testing stages, respectively. The area under the ROC curve (AUC) values of the DT and PCA-DT models in the testing phase are 0.8524 and 0.8345, respectively, while those of the NB and PCA-NB models are 0.8405 and 0.8779, respectively. The feature importance of the DT model shows that qc1Ncs is the most important feature, with a value of 0.45, followed by amax. The proposed ML models can help geotechnical engineers effectively assess susceptibility for preliminary design purposes.