OPTIMIZED MACHINE LEARNING METHODS FOR PREDICTION OF MARSHALL STABILITY VALUES

This study evaluates the performance of machine learning algorithms in predicting Marshall stability values to improve quality control processes in highway pavements. Coring is a costly, time-consuming and destructive method, which increases the need for alternative prediction models. In this context, Extra Trees, Random Forest, Gradient Boosting, K-Nearest Neighbours (KNN) and AdaBoost algorithms were used to predict the stability values obtained from core samples and error metrics were analyzed. In the study, the effects of hyperparameter optimization on model performance were examined in detail. The results show that the Extra Trees algorithm has the best prediction performance with an R² of 97.62% and an accuracy of 99.71%. Random Forest and Gradient Boosting algorithms also showed improvements after optimization, but their error rates remained higher compared to the Extra Trees model. The KNN model showed moderate success, while the AdaBoost model showed the lowest performance with an R² value of 58.87%. The findings reveal that machine learning algorithms can be used effectively in the prediction of stability values obtained from core samples and model performance can be improved by optimizing the right hyperparameters. The study shows that data-driven approaches can be less costly and time efficient in quality control processes.