Wed 15 Apr 2026 - Bridging the gap between models’ predictive power and interpretability is one of the key problems in modern predictive analytics specifically in insurance. Despite the availability of more performant machine learning (ML) tree-based models, less predictive GLMs are still a go-to method due to their explainable nature.
We propose a novel method for ensembling GLMs and GBMs and transform the state-of-the-art interpretability technique – SHAP. The resulting ensemble model, Interpretable Boosted GLM (IBLM), retains the linear formulaic representation and provides a set of per-observation parameter corrections. These corrections help modelers understand how the ensemble deviates from the underlying GLM while improving its performance.
The linear architecture of IBLM allows insurers to easily implement it into the existing rating structures, reducing or even eliminating friction costs of its implementation. The SHAP-corrected coefficients enable familiar interpretation of the rates for customers and stakeholders. Most importantly, the transparent nature of IBLM allows insurers better assessment of the risk they are exposed to.
Read the paper: Interpretable Boosted GLM, by Karol Gawlowski and Yafei (Patricia) Wang, FIA C.Act
Chair
Dylan is a qualified actuary, CERA, and data scientist. He began his career working in life insurance at HSBC, and then AIG Life. He now works in private medical insurance pricing as a data science actuary, building and deploying machine learning models.
At the IFoA he is the Deputy Chair of the AIDSET Research Sub-committee, with particular interests in eXplainable AI (XAI), privacy-preserving AI (notably federated learning), and AI fairness.
His research has been published in The Actuary magazine and he is currently developing a novel machine learning modelling method that ensures customer privacy can be met, while still maintaining high degrees of model accuracy.
Karol Gawlowski sits on the IFoA’s AIDSET Practice Board and chairs the Actuarial Data Science Working Party. He is an actuarial manager at EY UK, specialising in predictive modelling and machine learning for non-life insurance pricing and reserving, with an emphasis on interpretability and practical actuarial applications.
Patricia (Yafei) Wang is currently the Actuarial Director at Convex Insurance and the Deputy Chair of IFoA Actuarial Data Science Working Party.
| Members | Book for free |
| Non-members | Book for free |
Staple Inn
1-3 Staple Inn Hall
High Holborn
London
WC1V 7QH
The event will also be live streamed.