Ground investigations (GI) are essential prior to the design of construction projects. Among the different GI tasks, classifying soils into groups with similar properties is a fundamental geotechnical engineering process. Currently, experienced geotechnical engineers manually conduct soil classification using empirical tables based on laboratory or in-situ tests, which is labor-intensive and time-consuming. This study presents a machine learning (ML)-based approach to inferring soil types based on Cone Penetration Test (CPT) data. To identify an appropriate classification model, three classic algorithms, including Support Vector Machine (SVM), Artificial Neural Network (ANN), and Random Forest (RF), were built and validated on data collected from a reclamation project (The Project). Four important attributes from CPTs, including tip resistance qc, sleeve friction fs, pore-pressure u2, and depth d, were used as input features, and six soil types in The Project were applied as labels. The different models were compared based on their prediction performance and required learning time. The best results for both targets were obtained using the RF classifier, achieving over 90% accuracy.