Resilience-based Machine Learning Models for Restoring Interdependent Infrastructure Networks: Nodes Assessment Strategy for Post-Disruption Restoration

The resilience of critical infrastructure systems to disruptions is a paramount concern in contemporary society. To address this, the current study evaluates the effectiveness of several machine learning models for optimizing the restoration of interdependent infrastructure networks following disruptions to enhance their resilience. The models assessed include linear regression (LR), polynomial regression (PR), decision tree regressor (DT), K-nearest neighbors (KNN) regressor, and multi-Layer perceptron (MLP), with each model’s performance compared across multiple statistical metrics such as mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE), and R2 scores. Among the models evaluated, the MLP demonstrated the best performance with an R2 score of 0.962, a minimum RMSE of 0.187, and an MAE of 0.104, highlighting its capability to provide highly accurate predictions. In contrast, traditional models like Linear and Polynomial Regression showed comparatively lower performance, with R2 scores averaging 0.149 and 0.183, respectively, alongside higher error metrics. Decision Tree and KNN Regressors offered competitive results, particularly in minimizing prediction error, with RMSE values of 0.297 and 0.281, respectively. The findings suggest that advanced machine learning models such as MLP are more suitable for handling the complexities inherent in optimizing the restoration of interdependent infrastructure systems. These models offer better predictive accuracy and lower error rates, making them viable tools for decision-makers in infrastructure management. Future work may explore the scalability of these models across larger datasets and different disruption scenarios to further enhance their practical applicability.