The problem we have decided to solve is not having an accurate risk assessment of individual drivers before they get signed up for a car insurance policy. Currently, insurers ask for home address, age, gender, etc.
The solution we have come up with is adding another variable into the machine learning model, such as risk directly correlated with work address. A big portion of our driving is to and from work. Depending on the road we choose to take to work, our risk levels can vary widely.
The data sets we have used from Kaggle using insurance premiums predictions allowed us to predict the premium more accurately.
Therefore, we have two supervised machine learning models, their correlation between the road traveled is not very different due to lack of real data. We have used AWS Amazon’s Sage Maker and using their Jupyter Notebook using Python.
Conclusively, the insurance company should take this factor into consideration, there is about 20% accuracy and significant room for improvement, which could be achieved using real data.