Fuzzy-Rough Hierarchical Risk Assessment Algorithm And Its Application

Unreliable conclusions are often drawn in risk assessment tasks due to the existence of different types of uncertainty in the data. By introducing fuzzy-rough attribute weights to extreme learning machine (ELM) method, this paper presents a fuzzy-rough weighted ELM (FRW-ELM) algorithm in an effort to exploit the uncertainty information embedded in the data.This method proposes a fuzzy-rough hierarchical risk assessment approach that integrates FRW-ELM and the recently developed fuzzy boundary region-based fuzzy-rough feature selection (B-FRFS) technique to handle high-dimensional data effectively. In this risk assessment approach, the significance of each input node in the ELM is induced by its uncertainty degree contained within the boundary region of the associated fuzzy-rough set. Importantly, the required uncertainty measures of all the attributes are obtained as the by-product of the feature selection process that implements B-FRFS. In so doing, the resulting fuzzy-rough hierarchical risk assessment approach, consisting of B-FRFS and FRW-ELM, uses only B-FRFS-selected attributes as input nodes of FRW-ELM and the corresponding uncertainty degrees to compute their weights. Overall, this structure ensures the efficiency of the system. Systematic experimental results, for benchmark datasets and mammographic risk assessment problem, demonstrate that the proposed approach generally outperforms many state-of-the-art techniques in performing risk assessment tasks.