Risk Assessment In Homeowners And Renters Insurance Using Explainable AI Models

Abstract

Despite the increasing frequency and severity of extreme weather events and their consequences in the residential sector, decisions from homeowners and renters about their insurance coverage are generally insensitive to risk assessment, and pricing strategies are instead mostly based on availability bias and crude predictive methods. The development of sensible risk pricing relies on the estimation and visualization of risk levels for large populations, which entails the use of advanced big data methodologies. Due to the large number of factors affecting the risk of insurable assets, such estimations would only be feasible through explainable Artificial Intelligence approaches. This study proposes some explainable AI models, including Generalized Additive Models and Shapley Additive Explanations, to provide risk estimations and interpretations of risk levels for properties in Louisiana from several common perils. Each model offers its own intrinsic interpretation advantages, and all work together to facilitate the adoption of AI methods in an industry characteristically resistant to the use of big data techniques and low transparency levels. Insurers should use these tools in their[1] risk assessment and pricing strategies, benefiting from their potential to identify gaps in pricing, increase communication with clients, and improve their contribution to disaster prevention and mitigation interventions initiated by public policies. This contribution is particularly relevant in Louisiana, a state that has been affected by increasing damage from perils typical of the country’s hurricane climate during the last decades.