churn prediction using simple sklearn algorithms

My answers to questions: Question 1

1. Predicting customer churn using suprevised learning algoritms - seems to be the main purpose of this dataset
2. Customer segmentation and LTV analysis
3. Segmentation of sellers by unit price/total price/quantity/loyality

Question 2 Churn analysis seems to be the most important and valuable for the business. Building ML models, which predict customer churcn by their behaviour can help us to detect loyal clients and build an ongoing relationships with them for a long time. Also turn advertising to clients with high churn probability.

Question 3 see .ipynb file Validated accuracy for non-churn clients detection is 99,89% For churn clients 98,96% This was achieved by using Gradient Boosting classifier from sklearn package

Question 4 Dataset contains many categorial variables with simple encoding. So, application of logistic regression and knn algorithms failed to make predictive models. Practically, tree algorithms (as decision trees, random forest and Gradient Boosting) showed their efficiency with so-called "dummy encoded" variables. Also tree-containing algoritms have at least 2 tunable hyperparameters. We choosed maximum tree depth and leaf nodes to tune.

If I had more time, I would continue tuning of hyperparameters in GBC algorithm. But I also need more data :)