Multi-classifier information fusion in risk analysis

This paper develops a novel multi-classifier information fusion approach that integrates the probabilistic support vector machine (SVM) and the improved Dempster-Shafer (D-S) evidence theory to support risk analysis under uncertainty. Safety levels for various risk factors can be classified separately using the probabilistic SVM. Then, these multiple classification results will be fused at the decision level to achieve an overall risk evaluation by an improved D-S evidence theory with the integration of the Dempster' rule and the weighted average rule. The Monte Carlo simulation approach is employed to model the randomness and uncertainty underlying limited observations. A global sensitivity analysis is performed to identify the most significant factors contributing to the risk event. A realistic operational tunnel case in China is used to demonstrate the feasibility and effectiveness of the developed approach, aiming to assess the magnitude of the structural health risk. Results indicate the developed SVM-DS approach is capable of (1) Fusing multi-classifier information effectively from different SVM models with a high classification accuracy of 97.14%; (2) Performing a strong robustness to bias, which can achieve acceptable classification accuracy even under a 20% bias; and (3) Exhibiting a more outstanding classification performance (87.99% accuracy) than the single SVM model (63.84% accuracy) under a high bias (20%). Since the proposed reliable risk analysis method can efficiently fuse multi-sensory information with ubiquitous uncertainties, conflicts, and bias, it provides in-depth analysis for structural health status together with the most critical risk factors, and then proper remedial actions can be taken at an early stage.