Hybrid SVM–KNN Model for Soil Classification and Bearing Capacity Prediction in Geotechnical Applications

Abstract

Significant economic and ecological harm can result from harvesting operations that are not timed appropriately, especially when the number of vehicles involved exceeds the soil's holding capacity. This causes changes in nutritional and water conditions, compaction of the soil, and damage to tree roots and stems. The need for improved data on soil properties, particularly bearing capacity, is underscored by the fact that deep ruts created by vehicle movement further impede forest activities. First, the data was normalised for preprocessing in this study. Then, features were extracted using skewness, kurtosis, standard deviation, RMS, and crest factor. To forecast soil type and bearing capacity, the AdaBoost-SVM-KNN model was employed. This model optimises the parameters of SVM and adaptively modifies the parameters of KNN kernels in order to formulate component classifiers that are efficient. A weighted forecast was produced by averaging the predictions of the two models after the SVM updated the original KNN weights. An astounding 96.36% accuracy rate was shown by the results, proving that the AdaBoost-SVM-KNN model is capable of accurately soil classification and bearing capacity prediction. Better decision-making and the promotion of more sustainable forest management techniques could help reduce the negative effects of unsuitable harvesting activities.