Prediction of soil-bearing capacity on forest roads by statistical approaches

The soil-bearing capacity is one of the important criteria in dimensioning the superstructure. In Turkey, predictability of California Bearing Ratio values, which may be used in the planning and dimensioning of forest roads, of which about 26% lacks the superstructure, by using soil mechanical properties (cost and time efficient parameters that are easier to determine) is investigated. Simple linear regression, multiple linear regression, artificial neural networks and adaptive network-based fuzzy inference system methods were utilized. Two hundred sixty-four California Bearing Ratio values obtained from the project carried out on the forest roads of Bartin Forest Operation Directorate were used in both the production of training-test data and the creation of models. Statistical performance of the models was assessed by means of parameters such as root-mean-square error, mean absolute error and R². The obtained results show that the bearing capacity values predicted by artificial neural networks and adaptive network based fuzzy inference system models display significantly better performance than the simple linear regression and multiple linear regression models. While the highest prediction capacity belongs to adaptive network based fuzzy inference system (0.969-0.991), it is followed by artificial neural networks (R² = 0.796-0.974), multiple linear regression (R² = 0.796) and simple linear regression (R² = 0.554). What makes the algorithms superior than the traditional statistical models is the fact that they have many processing neurons, each with local connections, and thus have higher error tolerance. On the other hand, for the forest and rural roads, which play an important role in rural development of the forest peasants, to be able to operate all-seasons, superstructure should be immediately built in order to minimize the wear on the roads.