Machine Learning and AI in Property and Casualty insurance

In Property and Casualty Insurance, information is the currency that drives pricing, claim loss prediction and prevention, risk management and customer experience. AXA, a global insurance company, created a proof of concept (POC) using machine learning to optimize pricing by predicting “large-loss” traffic accidents with 78% accuracy.

Profitability in the insurance industry comes from two streams; the ability to identify high risks and then price them appropriately. Each year approximately 10% of AXA customers experience a loss. While the cost of most losses are in the hundreds or thousands of dollar range, about 1% are considered large losses; in excess of $10,000. AXA used existing information such as customer demographics and historical claim data, local and regional data, and other external data with machine learning to identify those customers who were likely to experience a loss in excess of $10,000 so that they could price it appropriately.

They began by identifying ~70 risk factors; driver age, address, vehicle type, prior loss history, vehicle age, original purchase channel …etc. The 70 risk factors were entered into a fully connected neural network with three hidden layers. AXA used existing data to train the model, and Cloud Machine Learning Engine’s HyperTune feature to tune hyperparameters.

The result:

• Large loss accuracy .0783
• Non large loss accuracy .0785
• Improvement over prior machine learning method (Random Forest) .0397
  

For all the details, read the blog