Case Study - Two Class Boosted Decision Tree

The objective of this study is to help insurance companies to take informed decision on the probability of person making claim based on the certain parameters recorded and/or processed while issuance of the policy.

Algorithm and Input Data:

Microsoft Azure “Two Class Boosted Decision Tree”, it is one of the most powerful ensemble methods which correct the errors in subsequent trees till the end.
Random data is generated for the following parameters which we assumed may have impact on the person health and could help in predicting the possibility of person making claim.

Parameters:
1. Age (Between 20- 50)
2. Gender (M/F)
3. Smoking(Yes/No)
4. Drinking (Yes/No)
5. Drugs(Yes/No)
6. Rash Driving Cases( Yes/No)
7. Disease Type-I (Life Threatening) –(Yes/No)
8. Disease Type-II – (Yes/No)
9. Life Threatening Job- (Yes/No)

While generating the data for the use of usage all data was generated as Boolean (0/1) for all parameters apart from Age. Following is the sample data set

Age	Gender	smoking	Drinking	rash driving- cases	Diseases Type 1	Diseases Type 2	Life threatening job	Claims Made
34	0(Female)	0(No)	0(No)	0(No)	0(No)	0(No)	0(No)	0(No)

Outcome:

In below example, after completing the Data modeling, Data Learning and Model Evaluation process of the algorithm, random data was entered to check the predictability of the algorithm.
We found that if a person with Drinking habit, rash driving cases and doing a life threating job have high probability of making a claim to insurance company.

Suggested Changes:

After demo following changes are suggested:
1. To make changes of some column type from numerical to categorical to simulate real world scenarios.
2. Generate Graphs for data presentation and display.
3. Need to increase the data input count and to see if performance improves.
4. A sample single page application.

Next Steps:

As this is one of the first and initial studies hence we need to improve upon usage and make it more usable to be used further.