Notes

One may ask how this model can be used by the client or account manager and it can be used greatly to predict high cost claims long before they are setup and processed.

For example, a look into the data will show that the peak claim cost for BigCo is around 200 days, you can use the model to anticipate high cost claims based on lets say head injuries and high narcotic presence. Thus being able to avoid that 200 day mark of claim setup and reducing cost and improving client health.

The model can be used in accordance with the narcotic addiciton program to get these type of at risk injuries in the program before the peak time of claim cost to reduce the chance of becoming a high cost claim.

In terms of model performance, it is always difficult to fit a dataset with such larger values and a large range. However, with an adjusted r squared value of.91 explaining the variation in the model and with similar RMSE and RSE values means the model fits the testing data well while not overfitting the training data.

ALso, it is important to consider ways in which a model can be improved, and I think maybe incorporating a split model based on subsidiary until the narcotic addiction program has set foot could improve performance simply because that 12% of claim cost that smallco brings will still create quite a bit of varaince for bigco.