Employing Data Mining Techniques for Engineering Soil Classification: A Unified Soil Classification System Approach

Ojeda J.M.P.; Huatangari L.Q.; Calderón B.A.C. · Mathematical Modelling of Engineering Problems (2023) · DOI: 10.18280/mmep.100609

Resumen

This research aimed to classify soils using data mining techniques in accordance with the unified soil classification system (USCS). Data were collated via certified soil mechanics studies and laboratory data collection sheets, providing a comprehensive matrix that reflects various soil properties as categorized by the USCS. The USCS system categorizes soils based on physical properties such as particle size and plasticity. The methodology employed was knowledge discovery in databases (KDD), executed in distinct phases. In the selection phase, influential variables for soil type were identified, with gravel, sand, plasticity index, and maximum dry density offering the most significant information. During the processing phase, anomalous and duplicate data were purged using various Python libraries. In the transformation phase, the variables were condensed by proposing five models. Lastly, during the data mining phase, classification algorithms such as k-nearest neighbors, support vector machine, decision trees, and random forest were implemented from the Scikit-learn library 1.1.2. These algorithms achieved accuracy levels of 68.22%, 73.83%, 84.11%, and 69.16% respectively. The decision trees algorithm, combined with the PL_03 model, predicted soil type with a reliability exceeding 84%. The implemented model is expected to assist engineers in making informed and precise decisions.

Instituciones: Universidad Nacional de Jaén