Hybrid Model: Teaching Learning-Based Optimization of Artificial Neural Network (TLBO-ANN) for the Prediction of Soil Permeability Coefficient

Abstract

The permeability coefficient (k-value) of the soil is an important parameter used in the civil engineering design of roads, tunnels, dams, and other structures. However, the determination of k-value by experimental methods in the laboratory or the field is still costly and time-consuming. Moreover, it requires special equipment and special care in the collection of soil samples for laboratory study. Therefore, in this study, we have proposed machine learning (ML) hybrid model: teaching learning-based optimization of artificial neural network (TLBO-ANN) to predict the k-value of soil based on limited parameters (natural water content, void ratio, specific gravity, liquid limit, plastic limit, and clay content) which can be determined easily in the laboratory. Test results of 84 soil samples obtained from the Da Nang-Quang Ngai expressway project in Vietnam are used in the model development. Statistical indicators such as correlation coefficient (R), root mean square error (RMSE), and mean absolute error (MAE) are used to validate and evaluate the accuracy of the model. The results show that the TLBO-ANN model is an effective tool in predicting correctly the k-value (R = 0.905) of soil for the consideration in the design of structures founded on the soil.

Predicting and Investigating the Permeability Coefficient of Soil with Aided Single Machine Learning Algorithm

“…Table 7 shows the comparison between the predicted permeability coefficient and the experimental permeability coefficient. Pham et al [33], Pham et al [34], Bui et al [35], and Ahmad et al [36] used the same number of 84 data samples for this investigation; however, the performance of the single ML model "gradient boosting" is significantly improved with R 2 � 0.971, RMSE � 0.199, MAE � 0.161, and MAPE � 0.185 for the testing dataset. With the ML model's sharply improved performance and reliability with R 2 (mean value of 10000 runs) � 0.804 for the testing dataset, the highest performance and the reliability of the GB model seem to come from the supplementary input variable "plasticity index" proposed in this study.…”

“…Although MCS computations take longer than K-fold CV computations, MCS findings are more dependable than K-fold CV results due to MCS's smaller variance, which is supported by the research of Fonseca-Delgado and Gomez-Gil [39]. Moreover, the effect of input variables on the predicted permeability coefficient of soil was not quantified in the investigations of [33][34][35].…”

“…is ML model required the input variable derived from the measurement of permeability coefficient; therefore, this model is limited in the practical engineering applications. Other ML models by Pham et al [33], Pham et al [34], Bui et al [35], and Ahmad et al [36] are proposed to predict the permeability coefficient of soil with six input variables including water content, void ratio, specific density, liquid limit, plastic limit, and clay content, which include the easy measurable properties of soil such as the Atterberg limits (liquid limit, plastic limit, and water content) and grain size distribution (clay content). Based on 84 data samples, Pham et al [33] revealed that the RF model could predict the permeability coefficient of soil with the performance metrics such as R 2 � 0.…”

“…Using the input variables as the easy measurable properties of soil makes the prediction of the permeability coefficient of soil by the ML model easier to access for engineers. However, the randomness was not taken into account in these ML models proposed in [33][34][35][36]. In fact, the reliability of these ML models was not verified by the validation technique such as K-fold crossvalidation or Monte Carlo simulation (MCS), which should be applied to enhance the performance and reliability of the ML model [18,37].…”

Application of hybrid modeling to predict California bearing ratio of soil

“…Machine learning (ML) development for solving real-world problems is receiving global attention in many fields. (Pham et al, 2016;Bui et al, 2022;Hadzima-Nyarko et al, 2022). They have recently been able to deal comprehensively with high nonlinearity and complex problems.…”