UNT Scholarly Works - 1 Matching Results
Search Results
- Prediction of Concrete Bridge Deck Condition Ratting Based on Climate Data in Addition to Bridge Data: Five States as a Case Study
- Evaluating the impact of learning from climate data, in addition to bridge data, on the performance of concrete deck condition rating prediction is critical for identifying the right data needed to enhance bridge maintenance decision making. Few studies have considered such an evaluation and utilized a small size of samples that prevent revealing the knowledge hidden within the big size of data. Although, such evaluation over big data seems quite necessary, class imbalance problem makes it challenging. To alleviate such a problem, five states, including Alabama, Iowa, New York, Pennsylvania, and South Carolina, were selected as the case study. Not only are the states located in three different climatically consistent regions defined by the National Ocean and Atmospheric Administration (NOAA), but also their concrete deck conditions ratings are somewhat balanced. To conduct the evaluation, this research developed the bridge data set pertaining to 56,288 bridges across the afore-mentioned states through employing the GIS technology. The bridge data set contains bridge data derived from National Bridge Inventory (NBI), and climate data derived from Parameter-elevation Relationships on Independent Slopes Model (PRISM) climate maps and NOAA. Then, two machine learning algorithms, including random forest and GBM, were trained - with and without climate data - and their prediction performances were compared. The results indicated that: (1) random forest outperforms GBM with an accuracy of 63.3%, and (2) the change in the prediction performance after further learning from climate data was marginal since the accuracy reached to 64.9%.