In the healthcare industry where medical insurance providers are competing with each other to acquire more and more customers, evaluating customers’ application to assign a risk level is of prime importance.
This helps in formulating the policies and the premium that a customer needs to pay. In order to work on this the insurance companies must share their data which is highly susceptible of being stolen and misused against them by their corporate rivals.
Federated learning (FL) works in a distributed fashion without sharing and accessing the actual data.
In this paper, we will try to analyze and exploit this concept to provide risk assumptions in the healthcare and life insurance industry.
The concept of FL is in a way a more decentralized and privacy protecting method of Machine learning. The basic problem it solves is allowing created models to earn knowledge from the data with actually accessing or exposing any section of it.
The centralized model then combines the entities received and shares it with the partners for the next iterations of local training. This type of FL is called cross-silo (where silo stands for organization).
The other type of FL is called cross-device. These devices can be computationally low-end such as smartphones. However, for some high data driven tasks FL avoids training on these low-capacity devices.