| Summary: | Today challenge concerning to transportation infrastructure networks is how to keep them operational under all conditions. Budgetary constraints and network dimension are between the main factors that make the management of a transportation network such a challenging task. Accordingly, aiming to support transportation network management tasks, a data-driven model for stability condition prediction of soil embankment slopes is proposed based on the well known Artificial Neural Networks (ANN) algorithm. For that, the ANN was feed with more than fifty visual features that usually are collected during routine inspections. The proposed model was addressed following two different strategies: nominal classification and regression. Moreover, to overcome the problem of imbalanced data, three training sampling approaches were explored: no resampling, SMOTE (Synthetic Minority Over-sampling TEchnique)and Oversampling. The main results are presented and discussed, comparing ANN predictive performance under the two strategies implemented. Also the effect of the three training sampling approaches is discussed. Moreover, aiming a better understanding of the proposed data-driven models, a detailed sensitivity analysis was applied, allowing to quantify the relative importance of each model input.
|
|---|