Stay or Go? Predicting Customer Churn through Graph Relationships
summary
A production pipeline models customers, products, tickets and events in Neo4j + GDS and extracts graph embeddings plus features from community detection, similarity, link prediction and centrality; those features feed XGBoost and MLP models and a gains table with SHAP explainers for evaluation. The system runs near‑real‑time ETL, highlights clusters of at‑risk users and reports a significant accuracy increase over tabular baselines, enabling targeted interventions and higher predicted customer lifetime value.
AI-written summary · about
chapters (10)
- 0:00Introduction to talk
- 1:07Why customer churn matters
- 4:18Limitations of traditional models
- 6:48Early warning signs of churn
- 9:14Graph-based approach with Neo4j
- 11:01Advanced churn prediction algorithms
- 16:17System architecture overview
- 19:09Future improvements and enhancements
- 22:01Outcomes and business benefits
- 24:28Q&A session
related talks
20mBuilding Intelligent Retention Engines: Real-Time Churn Detection
47mWhat Business Can Learn from Dating
18mGraph analytics: From zero to hero
32mGraph-based ML for Automated Health Care Services
31mBuilding a Customer-Intelligence Brain: How GraphRAG Turns Data into Decisions
34mModelling and predicting customer lifetime value
23mCustom-made games with machine learning and big data
06mData Science