Classification model on pet health insurance claims data with strong imbalance

I'm currently working on a project aimed at predicting pet insurance claims based on historical data. Our dataset includes 5 million rows, capturing both instances where claims were made (with a specific condition noted) and years without claims (indicated by a NULL condition). These conditions are grouped into 20 higher-level categories by domain experts. Along with that each breed is grouped into a higher-level grouping.

I am approaching this as a supervised learning problem in the same way found in …

classification classification model data datascience dataset health historical data instances insurance insurance claims null pet project