Toward data-driven operational risk management: Analysing the temporal effects of causal factors

Activity: Talk or presentationOral presentation

Description

Conference presentation at the OR64 conference at the University of Warwick, England on research investigating the temporal effects of causal factors to operational risks with a case study in aviation safety.

Additional information

Presentation abstract:

The consequences of inadequate or failed internal processes, systems and people – operational risks – are costly and disruptive for organisations globally. Understanding the factors contributing to such risks and their causal pathways is key to effective operational risk management. Current approaches predominantly involve individuals reviewing past incidents, investigating their causes and recommending corrective actions to minimise similar occurrences in the future. These manual and qualitative processes are (a) costly, (b) reactive, (c) biased by assessors’ subjectivities and experiences, (d) limited by human processing capacity in drawing inferences and similarities beyond a small sample of historical incidents, and (e) provide infrequent and static snapshots of the operational risk environment to decision makers.

Research is emerging using data analytics to identify what factors contribute to incidents, offering a more rigorous and consistent process for causal factors analysis. However, current data-driven approaches do not provide in-depth insights into how causal factors influence the probability of an incident. Further to the individual and collective effects of causal factors, the temporal effect between incident occurrences and causal factors has not been explored.

This study analyses how the temporal dynamics of a system can explain variations in the probability distributions of operational risk incidents. Using lagged terms, dynamic Bayesian networks are developed and empirically evaluated to model the relationships between causal factors and operational risk incidents over time. The optimal time-dependent model is subsequently compared with a baseline time-independent model, a static Bayesian network. The study is applied to aviation safety incidents, using five terabytes of one-secondly data on over one hundred dynamic sensors from the onboard flight monitoring systems of regional commercial aircraft.

While showcased with an aviation implementation, the enhanced data-driven approach to causal factors analysis is generalisable to other industries. Considering the temporal dynamics across an operational risk profile uncovers cumulative and delayed effects of causal factors on the probability of operational incidents. These insights indicate the speed that risky operating conditions can affect an organisation, thus enabling risk managers to efficiently allocate resources and implement mitigation strategies. More proactive management can reduce costly incidents, improve regulatory compliance and increase operational capacity.
Period14 Sep 2022
Event titleThe OR Society's Annual Conference: OR For a Better World Together
Event typeConference
Conference number64
LocationCoventry, United KingdomShow on map
Degree of RecognitionInternational