Machine learning-based soil classification: leveraging resistivity and CPT data for enhanced prediction accuracy

Conventional methods provide reliable values of soil properties critical for geotechnical design purposes, but they struggle to handle the complexities of geotechnical data effectively. The growing intricacy of soil properties necessitates a more precise and effective data-driven methodology in geotechnical engineering. Applying advanced methodologies, including machine learning and integrated data, is essential to address these constraints and enhance the accuracy and efficiency of analytical techniques. The study investigates the efficacy of machine learning in enhancing soil classification performance and evaluates the impact of integrating resistivity and CPT data. A detailed dataset incorporating electrical resistivity and key CPT parameters—cone resistance, sleeve friction, friction ratio, and total friction was compiled for model training and testing. Techniques for soil type classification employing machine learning algorithms, such as K-Nearest Neighbours, Random Forest, and Extreme Gradient Boosting. The assessment of the performance of each algorithm was based on some metrics, including accuracy, precision, recall, and F1-score. The study found that the machine learning algorithm effectively identified soil types such as poorly graded sand and silty sand. The integration of resistivity and CPT data led to a marked improvement in classification performance. Random Forest and XGBoost outperform KNN in soil type classification, with Random Forest achieving the best accuracy, precision, recall, and F1 score results. This work highlights the benefits of combining resistivity and CPT data in soil classification and demonstrates Random Forest and XGBoost’s superiority in handling intricate, multi-dimensional datasets. These findings suggest that this integrated approach can enhance the accuracy and efficiency of analytical techniques of geotechnical investigations.