Intelligent recommender system for car insurance plans - focusing on model flexibility and personalisation

Abstract

Acknowledging the success of personalized recommendations as support to promote sales within a business, this paper proposes the development of a recommender system to answer Fidelidade’s problem of depersonalization in the auto insurance sector. To build a model able to consider historical data from the customer and the car to recommend the best auto insurance package, a thorough data cleaning and model hypertunning were made to ensure that the three main objectives: predictability (accuracy when predicting), explainability (explaining to each customer the reason to recommend a certain product) and flexibility (proposing different coverages combinations) were satisfied.