Application of Artificial Neural Networks in Predicting Subbase CBR Values Using Soil Indices Data

Abstract

Subbase strength characteristics is one of the main inputs of pavement design, and such strength characteristics are normally represented by indices such as resilient modulus, dynamic modulus, and California Bearing Ratio (CBR), with the latter being a widely used index among pavement and geotechnical engineers. This paper examines the capability of Artificial Neural Networks (ANN) to develop a correlation between subbase CBR and primary soil data, which could help with estimating CBR for prediction purposes and with identifying the significance of each index with regard to subbase strength. Data were sampled from different areas in Karbala, Iraq, and a total of 358 subbase samples were used for model training and validation. The results showed that the proposed ANN model could successfully predict the CBR value using soil index data. Additionally, a sensitivity analysis was conducted to determine the importance of each contributing factor, and within the boundaries of the local subbase characteristics, the test results indicated that soluble salts were the most effective factor among soil parameters with an importance percentage of 39.46%, while the Plasticity Index (PI) was the least important factor, with a percentage of 2.06%. Based on the validity and quality of subbase soil tests, using ANN to predict CBR value may offer a suitable replacement for lengthy and expensive laboratory testing based on validated data for materials supplied from Karbala quarries.

On Random Subspace Optimization-Based Hybrid Computing Models Predicting the California Bearing Ratio of Soils

“…The prediction capability of the presented models, as was quantified through R 2 metrics, suggests the paradox of higher predictive accuracies based on predictive models developed using smaller datasets, compared to predictive models that used larger data sets and yielded comparatively lower R 2 values of moderate accuracy [49,50]. This is most probably a result of overfitting and network memorizing of the particular local dataset, which results in the models being weak in generalization.…”

“…In addition, Tenpe and Patel (2020) [49] used 389 soil test data and produced two models using SVM and GEP algorithms, with a performance accuracy ranging between 0.83 < R 2 < 0.90. Al-Busultan et al (2020) [50] used a dataset of 358 tests and developed an ANN model with an R 2 = 0.78.…”

Prediction of the Subgrade Soil California Bearing Ratio Using Machine Learning and Neuro-Fuzzy Inference System Techniques: A Sustainable Approach in Urban Infrastructure Development

“…Figure 11 The summary of the prediction metrics of the obtained model is compared in Table 6. The previous works in the literature showcased R 2 values ranging from 0.80 to 0.93 for MLRA, 0.78 to 0.97 for ANN, and 0.98 for ANFIS, respectively, for the CBR prediction with smaller datasets comprising 124 to 264 samples [25,26,[28][29][30][31]. In contrast, our investigation employed a more extensive dataset consisting of 2191 samples, encompassing diverse soil types.…”

Importance of artificial neural networks in civil engineering: a systematic review of the literature

“…Se han encontrado 11 estudios sobre la aplicación de las redes neuronales en la mecánica de suelos, en su mayoría para predecir las propiedades mecánicas de los suelos a partir de sus propiedades físicas. En [19] se aplicó un Per-ceptrón Multicapa para la predicción de la Relación de Soporte de California (CBR) de Subbase del pavimento usando datos de las propiedades físicas del suelo. En esta investigación se examina la capacidad de las RNA para desarrollar una correlación entre el CBR de la subbase y los datos primarios del suelo, lo que podría ayudar a estimar el CBR con fines de predicción y para identificar la importancia de cada propiedad física con respecto a la resistencia de la subbase.…”