The challenge
A niche insurance company operating in the medical and health domain needed to develop its pricing policies. To do so, it had to understand the risks associated with individual patients and adjust its pricing models accordingly, with the aim of achieving considerable savings.
Our solution
We built an application around a machine learning model that predicts the probability of a particular disease from many input features and parameters, including medical history. The model is a deep neural network trained on historical medical data covering five consecutive years, and it was designed to handle intrinsic difficulties in the data such as class imbalance.
We also built a validation framework to compare multiple approaches objectively and confirm that the chosen model clearly outperformed the alternatives.
- Disease probability prediction from medical history and other patient features
- Deep learning model trained on five consecutive years of historical medical data
- Handling of class imbalance in the training data
- Validation framework for objective comparison of competing approaches
The outcome
The model significantly outperformed the baseline models based on statistics. The insurer used its output to optimise pricing policy, increase revenue and manage risk better, which resulted in significant savings.
