Determination of Clay Content by Applying Machine Learning with Hydrometer Testing and Specific Gravity Analyses

บทคัดย่อ

This study aims to analyze and compare hydrometer test results with fundamental soil properties while applying Machine Learning (ML), a branch of Artificial Intelligence (AI), to enhance the speed and accuracy of clay content prediction. The study utilized soil samples from Nakhon Phanom and Sakon Nakhon provinces, Thailand. The experimental process included specific gravity and hydrometer analysis. For ML model development, linear regression (LR) and random forest regressor (RFR) were compared to analyzing factors influencing clay content. The data evaluation was based on feature importance analysis and statistical correlation (Correlation Matrix). The application of 10-fold cross-validation ensured that the models did not suffer from overfitting and confirmed the stability of predictions when using hydrometer data from longer test durations. The results indicate that hydrometer readings at longer durations exhibit a strong correlation with clay content and significantly improve the prediction accuracy of LR and RFR. The highest R² values obtained were 0.93 for LR and 0.87 for RFR, demonstrating that longer hydrometer test durations lead to more accurate clay content predictions. ML method combined with the hydrometer readings at 180 minutes, the R² exceeds 0.75. Specifically, LR outperformed RFR at minute 240, suggesting that the linear model better explains data variance at this duration. This research concludes that incorporating ML with hydrometer test data significantly improves the accuracy of clay content predictions. The findings highlight the potential of ML applications in soil property analysis and geotechnical engineering design, leading to more efficient and reliable engineering solutions.