| dc.description |
The Casagrande chart is traditionally used for determining soil classes. However, processing samples individually on this chart is time-consuming, and human error, particularly at the classification boundaries, can lead to incorrect soil classification. To address these issues, this study employs machine learning algorithms to classify different soil types more efficiently and accurately. The primary goal is to integrate machine learning into engineering geology studies, leveraging technological advancements. As part of the study, field and experimental work was conducted, beginning with the collection of 272 soil samples from the designated study area to represent the entire region. The initial physical properties of these samples were then determined. The soil samples were carefully double-bagged and transported to the laboratory to prevent any degradation. Upon arrival, the water content of the samples was determined first. Subsequently, sieve analysis, consistency limits, and specific gravity tests were conducted in sequence. For the classification of soil types, machine learning methods, including Decision Tree (DT), Support Vector Machines (SVM), K-Nearest Neighbor (k-NN), and Artificial Neural Networks (ANN), were employed. Among these, the DT model demonstrated the highest performance, achieving a success rate of 91.5% across five different soil classifications. The ANN, SVM, and k-NN models followed, with success rates of 89.0%, 89.0%, and 86.0%, respectively. In addition, the hyperparameter optimisation of the models used in the study was provided and the minimum classification error analysis was presented. This study significantly contributes to the effective and efficient analysis of experimental data, demonstrating that soil classification can be successfully performed by integrating machine learning methods with numerical data. |
|