Customer-Churn-Prediction

Developing a rolling-window churn prediction system with Python and SQLite

Objective

▪︎ To Create and evaluate a churn prediction system so that each week the retailer can predict customers who are likely to churn and take preventive action.

▪︎ To provide insights into what differentiates people who churn vs. those that stay

Dataset

▪︎ Transactional database in Sqlite

▪︎ The database contains 64,228 transaction records of 5,359 customers in six months

▪︎ Based on preceding research, customer churn is defined as "31 days of inactivity"

What's included in the project

▪︎ Extract data from sqlite3

▪︎ Feature Engineering including building a time-series dataframe in sqlite3 and storing the results in Pythonpandas

▪︎ Investigation of Feature Corrlation using Python seaborn heatmap

▪︎ Data Standardization

▪︎ Building Machine learning models: svc, random foreast

▪︎ Parameter tuning & Model Evaluation with Pipeline

▪︎ Model deployment & Prediction

▪︎ Customer Pen Portraits

Result Preview