JOB-A-THON - January 2023

Problem Statement:

To predict the CLTV based on the user and policy data of one of the leading insurance companies in India (VahanBima).

Given data:

Input Variables:
    Id - Unique identifier of a customer
    Gender - Gender of the customer
    Area - Area of the customer
    Qualification - Highest Qualification of the customer
    Income - Income earned in a year (in rupees)
    marital_status - Marital Status of the customer {0: Single, 1: Married}
    Vintage - No. of years since the first policy date
    claim_amount - Total Amount Claimed by the customer (in rupees)
    num_policies - Total no. of policies issued by the customer
    Policy- Active policy of the customer
    type_of_policy - Type of active policy  
Target Variable:
    Cltv - Customer lifetime value

Solution:

Approach-1:

• Preprocessing (EDA/Data Engineering):
  • Checked missing values (None Found).
  • Converted the categorical data into nominal and ordinal values using MsExcel.
  • Scaled the numerical data (claim_amount & cltv) using sklearn MinMax().
  • Checked relationships between input and target variable (Correlation)
• Processing (Model Creation and training/validating):
  • Looped through Regression models (RandomForestRegressor, GradientBoostingRegressor, DecisionTreeRegressor, LinearRegression, Lasso, Ridge) to get the best r2_score possible with the above pre-processed data – Found GradientBoostingRegressor performed well in comparison to all the models with a r2_score of 0.1603.

Approach-2:

• Based on the above preprocessed data, a deep neural network is trained to see if it can be used to solve the problem statement.

Approach-3:

• Used CatBoostRegressor wherein preprocessing steps like conversion and scaling are not required as it is capable of giving quick results without the need of any preprocessing steps to save time.