The potential of nonparametric model in foundation bearing capacity prediction

Nonparametric mathematical models have gained a very massive attention in the last two decades in solving regression problem. The application of soft computing methodologies produced a very remarkable assistance to human abilities especially in solving nonlinear and non-stationary engineering problems. The current article investigates the utility of k-nearest neighbor (k-nn) approach in predicting ultimate bearing capacity of shallow foundation. The inspected application involves an experimental data set of foundation dimension and soil properties that suggested and calculated via manual computational methods. The predictive model is established using dimensional shallow foundation, and soil properties are an inputs variable, whereas the bearing capacity is the output variable. For the purpose of comparison and evaluating the modeling accuracy, multiple linear regression (MLR) model is chosen to diagnose the result accuracies. Couple of statistical indicators are utilized to exhibit the performance criteria of the predictive model including coefficient of determination (r(2)), degree of agreement (d), root-mean-square error (RMSE) and mean absolute percentage error (MAPE). The results exhibited a very representable and high accuracies of the investigated k-nn model vis-a-vis MLR. For instance, the RMSE and MAPE were enhanced by 24 and 17%, respectively. In addition, the findings indicated that k-nn provides an accurate and reliable alternative predictive model to the manual computational methods.