M5 Soft Computing Techniques for Assessment of Soil Liquefaction

Sayed, Abu; Rahman, Md Mahabub

doi:10.22075/jrce.2025.34669.2134

	M5 Soft Computing Techniques for Assessment of Soil Liquefaction
Journal of Rehabilitation in Civil Engineering
مقاله 23، دوره 13، شماره 3 - شماره پیاپی 39، پاییز 2025، صفحه 199-214 اصل مقاله (1.04 M)
نوع مقاله: Regular Paper
شناسه دیجیتال (DOI): 10.22075/jrce.2025.34669.2134
نویسندگان
Abu Sayed¹؛ Md Mahabub Rahman^* ²
¹Lecturer, Department of Civil Engineering, Pundra University of Science and Technology, Bangladesh
²Lecturer, Department of Civil Engineering, Hajee Mohammad Danesh Science and Technology University, Bangladesh
چکیده
It is essential to precisely estimate the liquefaction potential because soil liquefaction is a factor that raises the quantity and intensity of losses in an earthquake. In the past, the prediction of soil liquefaction was based on multiple analytical inferences. The purpose of this research is to develop an M5 model for both classification and regression in order to investigate the suitability of the M5 decision tree for liquefaction assessment. Additionally, the divisional approaches of fuzzy clustering means (FCM), kfold clustering, and grid search cross-validation (Gridsearch CV) are investigated in order to create effective regression and classification models. In this work, specific models are developed using a data set of 200 boreholes from standard penetration tests on soils in the Dinajpur region. The efficacy of the constructed models is assessed using several performance measures, such as root mean squared error (RMSE), mean absolute error (MAE), and coefficient of correlation (R) for regression models, and accuracy, precision, and AUC value for classification models. Based on the results, it was found that the M5 decision tree regression model shows R = 0.95, IoA = 0.86, and IoS = 0.96 for testing and R = 0.93, IoA = 0.88, and IoS = 0.96 for training data. On the other hand, the classification model shows accuracy = 95%, recall = 1, and F1 score = 0.97 for testing and 98.75%, 1, and 0.99 for training, respectively. Both of these results were found for the Kfold technique, which predicts a more accurate value than other divisional approaches.
تازه های تحقیق
This work aims to examine the applicability of M5 decision tree machine learning algorithms to predict liquefaction vulnerability. Both classification and regression models have been developed to determine nonlinear relationships between the physical properties of the soil. In this study, 200 data sets were used for different divisional approaches such as grid search cross validation (Gridsearch Cv), kfold clustering means, and fuzzy clustering means (FCM) to develop successful regression and classification models. It was found that the developed M5 decision tree regression model shows R = 0.9471 for testing and R = 0.9297 for training. The developed M5 decision tree classification model shows an accuracy of 95% for testing and 98.75% for training. The Kfold cross-validation technique predicted a more accurate value compared to other divisional approaches.
کلیدواژه‌ها
Soil liquefaction؛ Machine learning techniques؛ Regression model؛ Classification model؛ Performance metrics
مراجع

آمار تعداد مشاهده مقاله: 494 تعداد دریافت فایل اصل مقاله: 384

دانشگاه سمنان

پیوندهای مفید

اداره چاپ و انتشارات دانشگاه سمنان

آمار

تعداد نشریات	22
تعداد شماره‌ها	718
تعداد مقالات	10,319
تعداد مشاهده مقاله	72,323,118
تعداد دریافت فایل اصل مقاله	64,046,365