1 of 9

CUSTOMER CHURN PREDICTION IN INSURANCE COMPANY

GLORIE METSA WOWO

​

DATA ENTHUSIAST

—------------------------------------------------------------------------

Saturday, 29th October 2022

2 of 9

Objectives

Could policy data about a specific customer possibly show the pattern of his/her churn ?

Problem statement

  1. Understanding how to use machine learning in insurance for churn rate

​

  1. What are the parameters needed

​

  1. Which models to use

​

  1. Key Recommendations

​

​

3 of 9

Light Gradient Boosting Method (LightGBM)

Multilayer perceptron

LOGISTIC REGRESSION

EXAMPLE OF MODELS

—-------------------

4 of 9

Shape before cleaning : 100,000 rows, 31 columns

Shape after cleaning : 86000 rows, 18 columns

5 of 9

Feature engineering

  • Create new features
  • Encode data (one hot encoding, label encoding)
  • Scale data
  • Balanced data (SMOTE)

Shape of data after cleaning : 86000 rows, 18 columns

Split data : train : 75 %

Validation : 25 %

6 of 9

LR

MLP

LGBM

Performance on Train/validation

7 of 9

8 of 9

KEY - RECOMMENDATION

  • On premium (PRIMTOTA) :

​

Higher premiums might encourage customers to shop around to find a better deal elsewhere.

  • On location (BUREAU) :

The subscription office (BUREAU),

​

linked to the geographical

position of the customer ?

​

Churn rate could be influenced by the migration of the population who cannot get insurance because of the distance with their insurer.

  • And other :
    • on on customer’s occupation (PROFESSION),
    • the type of subscription,
    • the duration of the contract validity as relevant features.
    • The month of the cancellation

9 of 9

KEY - RECOMMENDATION

Collect data efficiently if you want to use it to understand your market and take more accurate decision

LINK TO DEMO