words per minute
6
MukeshKumarChaudhary
00:00
Speed
You have asked us to build a model that can predict if an applicant will be a high-value or low-value customer. We were supplied with 48,842 observations on prior customers that can be used to build a model and gauge the potential profit from using it. I was informed that this product would be sold only to customers age 25 and older, so younger policyholders were removed from the data, leaving 40,410 to analyze. The model we constructed is relatively simple to apply and produced a projected profit of 4.92 per applicant. Prior to building the model we checked the data in a variety of ways. The data contained the applicant’s age, education level, marital status, occupation group, capital gains, the number of hours worked per week, a proprietary insurance score developed by MEB, and an indicator of whether the applicant was high or low value. Other than for occupation group, there were no missing values. In examining the data it appears that all the variables may be useful in predicting customer value. In particular, high value tends to be associated with older ages, more years of education, being married with spouse present, being in a high-numbered occupation group, working more hours per week, having a higher insurance score, and having more capital gains. I also noted that there were two types of married individuals. One had very few observations and a similar relationship to being high value. Those two were combined into a single category of being married with spouse present. I then tried a variety of models to see which would perform best. This is done by calibrating a model on 70% of the data and then seeing how that model performs on the other 30%. This replicates the way our model will be used