Leveraging dependencies

Inspectors often check a collections of specific items to verify whether an organisation is compliant or not. Utilizing dependencies between these check can strengthen models and support inspectors during inspections.

Inspectorates are vital to our society to ensure that companies, institutions and people operate safely and adhere to laws and regulations. Because they only have limited resources, inspectorates cannot inspect every aspect of every company or institution. Currently risk modelling is usually done by aggregating all factors into one risk score. However many of those factors are related and modelling at a more-detailed level can help inspectors. For example: if the floors are not clean, it may be likely that the kitchen is not clean either. If we find these dependencies, we may be able to create a model that can recommend aspects to check for an inspector, based on things that have already been inspected. In this ways inspectors can dynamically prioritize their inspections.

Key Challenges in this domain

Data in the inspection domain is often sparse on the aggregated level. Modelling in more detail on the different aspects of compliance increases the challenge of sparse data. To find dependencies between different factors in an inspection, expert knowledge is important to identify how factors relate to each other. Finally, these connections need to be modelled in such a way that they model can provide concrete and useful suggestions to inspectors.

Key Questions

  • How can we extract dependencies between different factors of compliance?

  • How can we model the dependencies with expert knowledge and sparse data?

  • How can we best support the inspector during their inspections with this fine grained knowledge?

Solutions

This work aims to advance risk classification beyond binary labels by learning interdependencies between inspection items using probabilistic graphical models like Bayesian networks. These models aim to support interactive inspections by prioritizing items dynamically, combining data-driven learning with expert knowledge to handle incomplete information effectively.

Meet the researcher

Sebastiaan Jans

Utrecht University

""

Results

Inspectorate Use Cases

Description of use cases that have been executed within this work packages

Publications

Check out the publications related to Leveraging Dependencies