Acte I · The context
1. e4 e5
The dataset is a serious classic: Telco Churn, 7,043 customers of a telecom operator, 35 columns that describe each of them (subscribed services, contract type, tenure, charges, payment method), and one label: left, or stayed.
Acte II · The problem
2. Nf3 Nc6
A customer who leaves does not give notice. By the time the operator notices, it is already too late. The question: can the early signals of a departure be read in the data, and turned into a reliable prediction?
Acte III · The approach
3. Bc4 Bc5
Exploration first: 12 bivariate analyses to understand who leaves and why, before any modelling. The clearest signals:
- Contract type dominates everything: month-to-month contracts concentrate the highest churn.
- Tenure protects: the highest risk sits in the first 12 months.
- Add-on services retain: tech support, online security, backup and device protection are all associated with fewer departures.
- The payment method speaks: paying by electronic check is associated with maximum churn.
- Monthly and total charges complete the picture.
Section in preparation
Acte IV · The results
4. O-O
Section in preparation
Acte V · What I learned
5. d4
Section in preparation