Sairamr46/price-increase-churn-model

Price Increase Churn Prediction Model

Problem

Predict whether a subscription customer will cancel within 90 days after a price increase.

Model Architecture

• Algorithm: XGBoost Gradient Boosted Trees
• Type: Binary Classification (Stay vs Churn)
• Target: Churn within 90-day window after price increase

Dataset

• Source: aai510-group1/telco-customer-churn (IBM Telco Customer Churn)
• Size: 7,043 customers
• Churn Rate: 26.5%

Feature Engineering

Core Features (mapped from your data)

Your Column	Engineered Features
tenure	tenure, tenure_log, tenure_squared, tenure_group
days_left_in_contract	days_left_in_contract, contract_ending_soon, is_month_to_month
rmr (Monthly Charge)	rmr, rmr_log, rmr_per_tenure_month, new_monthly_charge, monthly_increase_amount
annual_services	num_services, service_diversity, rmr_per_service
last_appointment_date	satisfaction_score, churn_risk_score, low_satisfaction
added_date	age_group, cohort proxies

Price-Increase-Specific Features

• price_increase_pct — configurable (default 15%)
• new_monthly_charge — post-increase monthly cost
• monthly_increase_amount — absolute dollar increase
• price_to_cltv_ratio — price burden relative to lifetime value
• price_increase_burden — increase relative to historical spend
• Interaction terms: tenure × price_increase, contract_ending × price_increase, low_satisfaction × price_increase

RFM-Style Features

• recency_score — from satisfaction (lower = higher recency)
• frequency_score — from service count quintiles
• monetary_score — from total revenue quintiles
• rfm_score — combined RFM score

Performance

Metric	Value
AUC-ROC	0.9987
Average Precision	0.9968
Precision	0.9733
Recall	0.9759
F1 Score	0.9746

Files

File	Purpose
train.py	Main training script
predict.py	Inference script with sample data
model.py	Model class with evaluation
features.py	Feature engineering pipeline
data_loader.py	Dataset loading
config.py	Hyperparameters
price_increase_churn_model.pkl	Saved model artifact

Usage

Training

cd /app/price_increase_model
python train.py

Inference

from predict import predict_churn_risk
import pandas as pd

# Your customer data
customers = pd.DataFrame({...})
results = predict_churn_risk(customers, 
                              model_path='price_increase_churn_model.pkl',
                              price_increase_pct=0.15)
print(results)

Risk Tiers

Probability	Tier	Action
≥ 0.70	High Risk	Immediate retention outreach
0.40 - 0.70	Medium Risk	Targeted offer / discount
< 0.40	Low Risk	Standard communication

How to Adapt to Your Data

1. Map your columns to the feature engineer in features.py
2. Set your price increase % in config.py or pass to inference
3. Add your actual price increase date to compute days_since_price_increase
4. Replace Churn target with your actual 90-day cancellation flag
5. Retrain on your historical data

Business Impact

This model enables:

• Proactive retention: Identify at-risk customers before they churn
• Targeted discounts: Offer retention incentives only to high-risk customers
• Price optimization: Simulate churn impact at different increase levels
• CLV protection: Prioritize retention of high-value at-risk customers

Generated by ML Intern

This model repository was generated by ML Intern, an agent for machine learning research and development on the Hugging Face Hub.

• Try ML Intern: https://smolagents-ml-intern.hf.space
• Source code: https://github.com/huggingface/ml-intern