Self healing databases for predictive risk analytics in safety-critical systems

Abstract Assuring the quality, consistency and accuracy of safety data repositories is essential in safety-critical systems. In many systems, however, significant effort is required to identify, address, clean and repair data errors and inconsistencies, and to integrate safety data sets and repositories, particularly for risk analyses. Although some self healing and self repairing capabilities leveraging machine learning and predictive analyses have been employed to identify anomalies and monitor quality in structured safety-critical data sets, little attention has been focused on addressing shortcomings in heterogeneous—structured and unstructured—safety data sets, the focus of this work. The text mining and classification analysis employed in this research indicates that machine learning techniques can be employed to improve the accuracy and robustness of large-scale structured and unstructured database repositories, and to improve the effectiveness and efficiency of safety data repository maintenance. Hybrid machine learning approaches, leveraging machine learning, text mining and natural language processing, offer additional promise in future work.