Development of a Streamlit-Based Deep Learning Tool for Instant Soil Classification from Borehole Grain Size Data

Soil classification is an important part of geology in geotechnical engineering, because it affects the design of foundations, slope stability, and the safety of the construction site. This study presents an easy, dependable, and intelligent soil classification framework using a Multilayer Perceptron (MLP) deep learning model. Data used to train the MLP model included both real borehole grain size distributions and synthetic granular soil data, with synthetically generated data used due to the limitations of previously small datasets in terms of size. Inputs included percentages of gravel, sand, and fines, metrics for grain size such as D10, D30, D50, and D60, and Gs. The MLP model was developed to classify soil according to the Unified Soil Classification System (USCS).MLP model training was monitored using a loss curve, while performance evaluation utilized a confusion matrix, with precision, recall, and F1-score metrics being evaluated on a class-by-class basis so the assessments of classification accuracy can be robust. The proposed classification method showed high performance in soil classification during the entire USCS, thus offering geotechnical engineers an alternative to the slow, manual soil classification techniques that may be fallible due to human error.To improve ease of access and use, a website based platform through Streamlit was developed to allow geotechnical engineers to input grain size data, obtain soil types, and visualize performance in real time. This tool is designed to eliminate mistakes, allow for fast analysis, and advance data-driven decisions in geotechnical investigations.