Prediction of Road Maintenance and Repair Costs in Turkey Using a Hybrid Modelling Approach

Abstract

Accurate estimation of road maintenance and repair costs is of strategic importance for the efficient management of public resources and the safety of transportation systems. In Turkey, these costs are influenced by a wide range of multidimensional factors, including meteorological and environmental conditions, infrastructure characteristics, traffic intensity, economic indicators, and financial cost components. This study aims to comprehensively examine these factors and to develop a high-accuracy prediction model for road maintenance and repair costs. A national dataset covering a 19-year period (2004–2022) and comprising 21 independent variables identified through the literature and expert judgement was employed. Methodologically, classical statistical approaches – Multiple Linear Regression, Ridge Regression, Least Absolute Shrinkage and Selection Operator, Stepwise Akaike Information Criterion, and Granger Causality Analysis – were integrated with soft computing techniques, including Random Forest, Gradient Boosting, Support Vector Machines, Artificial Neural Networks, Genetic Algorithms, Principal Component Analysis, and Sensitivity Analysis. In total, ten variable-selection techniques were combined with five prediction models, resulting in 50 hybrid model configurations. The results indicate that the RR-ANN hybrid model, constructed using Freight KM, bitumen and salt consumption and minimum wage variables selected via RR, achieves the highest predictive accuracy (MSE = 175 474.92; RMSE = 418.90; R² = 0.985; AdjR² = 0.980; MAPE = 1.36%). Computational performance analysis further shows that the trade-off between accuracy and execution time is critical in practical applications: RR-based models are the fastest, whereas ANN-based hybrids provide superior accuracy at the cost of higher computational and implementation effort. Overall, the findings demonstrate that hybrid modelling approaches yield more reliable and robust predictions than single-method specifications. The proposed framework contributes methodologically to the literature and offers policymakers a practical and standardised tool to support budget planning and the development of sustainable road maintenance strategies.