AI-Driven Automation and Digital Twin Framework for Smart Transportation Systems and Predictive Infrastructure Maintenance

Abstract

With the accelerated pace of urbanisation and growing transport demand, traffic congestion has worsened, while transportation infrastructure has increasingly deteriorated, particularly in rapidly growing metropolises such as Lagos, Nigeria. Traditional transportation management systems rely on reactive maintenance strategies and static traffic-control mechanisms, resulting in higher operational costs, infrastructure-related challenges, and reduced mobility efficiency. This study proposes an AI-driven automation and digital twin framework for smart transportation systems that leverages real-time sensor data, historical transportation information, Internet of Things (IoT) technologies, and advanced machine-learning algorithms for intelligent traffic management and predictive infrastructure maintenance. The framework combines real-time digital twin technology with AI-based predictive analytics to address both traffic optimisation and infrastructure-deterioration prediction within a single decision-support platform. It comprises data acquisition, data pre-processing, feature engineering, machine-learning model optimisation, and digital twin-based decision support for continuous monitoring and predictive analytics. Seven machine-learning algorithms—Linear Regression, Random Forest, XGBoost, LightGBM, CatBoost, Support Vector Regression (SVR), and Multi-Layer Perceptron (MLP)—were evaluated using an 80:20 training-to-testing split, with hyperparameter tuning conducted through Grid Search and Randomised Search. Model performance was assessed using the coefficient of determination (), Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE). Evaluation over 30 independent runs showed that XGBoost achieved the best traffic-congestion prediction performance, with an  of 97.3% ± 0.9%, an MAE of 1.62 ± 0.14, an RMSE of 2.21 ± 0.18, and a MAPE of 4.1% ± 0.6%. For predictive infrastructure maintenance, Random Forest achieved the best performance, with an  of 96.8% ± 1.0%, an MAE of 1.47 ± 0.13, an RMSE of 2.05 ± 0.17, and a MAPE of 3.8% ± 0.5%. Statistical significance analyses supported the selection of the best-performing models, although some pairwise differences were not significant. The proposed framework enables accurate real-time traffic prediction, proactive maintenance scheduling, and intelligent transportation management, thereby improving urban mobility, enhancing infrastructure reliability, reducing maintenance costs, and supporting sustainable transportation planning. The findings demonstrate the potential value of integrating artificial intelligence and digital twin technologies for next-generation smart transportation systems and predictive infrastructure maintenance.