decisiontree_predictive_modelling

Problem Statement

An insurance provider (US based) offers health insurance to customers. The provider assigns a PCP (primary care physician) to each customer. The PCP addresses most health concerns of the customers assigned to them. For various reasons, customers want change of PCP. It involves significant effort for the provider whenever the customer makes a change of PCP.

You will find a subset of the insurance provider data along with PCP changes. The provider likes to understand why are members likely to leave the recommended provider. Further, they like to recommend a provider to them that they are less likely to leave.

The dataset consists of following fields:

Columns	Description
Id	Column identification field
Outcome	Member changed to his/her preferred primary care provider instead of auto assigned to. 0: Member keeps the auto assigned provider. 1: Member changed to this provider by calling customer service.
Distance	Distance between member and provider in miles
Visit_count	Number of claims between member and provider
Claims_days_away	Days between member changed to/assigned to the provider and latest claim between member and provider
Tier	Provider Tier from service, values - 1, 2, 3, 4. Tier 1 is highest benefit level and most cost-effective level
Fqhc	Value 0 or 1 1: Provider is a certified Federally Qualified Health Center
Pcp_lookback	Value 0 or 1 1: The provider was the member's primary care provider before
Family_Assignment	Value 0 or 1 1: The provider is the pcp of the member in the same family
Kid	Value 0 or 1 1: Member is a kid. (under 18 for state of New York)
Is_Ped	Value 0 or 1 1: Provider is a pediatrician
Same_gender	Value 0 or 1 1: Provider and member are the same gender
Same_language	Value 0 or 1 1: Provider and member speak the same language
Same_address	Value 0 or 1 1: The re-assigned provider has the same address as the provider pre-assigned

Aim:

• Build a Predictive Model
• Evaluate the model
• Refine the model, as appropriate

Perform the following tasks:

• Select a method for performing analytics
• Preprocess the data to enhance quality
• Carry out descriptive summarization of data and make observations
• Identify relevant, irrelevant attributes for building model
• Perform appropriate data transformations with justifications
• Generate new features if needed
• Carry out the chosen analytic task. Show results including intermediate results, as needed
• Evaluate the solutions
• Look for refinement opportunities

Inferences

Objective

An health insurance provider (US based) offers health insurance to customers. The provider assigns a PCP (primary care physician) to each customer. The PCP addresses most health concerns of the customers assigned to them. For various reasons, customers want change of PCP. It involves significant effort for the provider whenever the customer makes a change of PCP. You will find a subset of the insurance provider data along with PCP changes. The provider likes to understand why members are likely to leave the recommended provider. Further, they like to recommend a provider to them that they are less likely to leave.

Exploratory Data analysis (EDA)

• Read the data from “DataSet_PCP_Change.csv” csv file

• Data Summary:

  • Summary of Healthcare Insurance Enrolled Members (Auto PCP/Switched PCP) – Total 3,130 members have enrolled to Healthcare insurance benefits, of that 127 (4%) members have switched from assigned Auto PCP and interestingly 96% members are on Auto PCP

    	Enrolled Members
    Category	Count	Percentage
    Auto PCP	3,003	96%
    Switched PCP	127	4%
    Total	3130	100%

  • Summary of Gender - Enrolled Members (127 Members) Who Switched PCP – Total 127 members have switched PCP, of that 62 (48%) are kids and 65 (52%) are adults

    	Switched PCP Members
    Gender	Count	Percentage
    Kids	62	48%
    Adults	65	52%
    Total	127	100%

  • Summary of Pediatric PCP/Non-Pediatric PCP Vs Switched PCP Members (127 Members) – There are 127 members who have switched PCP (pediatric and non-pediatric), of that 12% of members among Tier-1, 13% of members among Tier-2, 6% of members among Tier-3 are treated by non-pediatrician (assumption here being switched PCP members are not assigned the appropriate physician, so there's high possibility that they will request further to get new PCP assigned. Therefore, here there is an opportunity to improve the model)

    Tier		Tier-1		Tier-2		Tier-3		Tier-4
    Pediatric/Non-Pediatric	Adult/Kid	Count	Percentage	Count	Percentage	Count	Percentage	Count	Percentage
    Non-Pediatric PCP	Adult	35	54%	8	27%	10	63%	11	69%
    Non-Pediatric PCP	Kid	8	12%	4	13%	1	6%	-	0%
    Pediatric PCP	Adult	-	0%	-	0%	1	6%	-	0%
    Pediatric PCP	Kid	22	34%	18	60%	4	25%	5	31%
    Total		65	100%	30	100%	16	100%	16	100%

Model Building

• Used the Decision Tree Regressor to recommend the appropriate PCP to members - 30% of PCP Dataset was used for testing and 70% of PCP Dataset was used for training

• Achieved a model accuracy of 96%