why_Drop.txt

1. Customer ID

Reason: Unique identifier. Does not contain any pattern that helps predict churn.

Customer ID → unique key, no predictive value

2. Geolocation fields

Reason: Too granular for churn prediction in this dataset. Adds noise, no business logic benefit.

Country → same region for most customers, no variation  
State → too granular, no churn pattern  
City → unnecessary location granularity  
Zip Code → location noise, not behavior  
Latitude → raw coordinates, useless for churn logic  
Longitude → same as above  
Population → not linked to individual churn behavior

3. Post-churn leakage columns

Reason: These fields reveal churn after it already happened, making the model cheat. Must remove.

Customer Status → directly tells churn outcome  
Churn Score → calculated after churn event  
Churn Category → reason label after churn confirmed  
Churn Reason → tells exactly why they churned (leakage)

4. Some more columns
Quarter → only one value
Total Revenue → similar meaning to Total Charges (redundant)
Under 30 → derived from Age, redundant feature
Number of Dependents → similar info to Dependents yes/no column