Unlocking Churn Prediction with Graph Neural Networks
Learning objective
Unlocking Churn Prediction with Graph Neural Networks
Learning objective
Learning Objective
At the end of the lesson, you will understand how GNNs enhance churn prediction, explore customer interactions, and create insights using Power BI.
What do you already know about customer churn prediction and Graph Neural Networks?
Project Overview
Explore how GNNs tackle high churn in telecom, focusing on customer-customer interactions. Visual insights will aid strategy.
Research Question & Objectives
Can GNNs improve churn prediction and offer insights via visual tools? Objectives: Evaluate methods, develop frameworks, create dashboards.
Literature Gap
Current models excel in performance but lack relationship modelling. GNNs and visualisation are underexplored.
Methodology Snapshot
Dataset: 7,043 records, 45 features. Nodes: customers, Edges: cosine similarity. Tools: networkx, scikit-learn, PyTorch Geometric.
Models Tried
GCN as baseline, GraphSAGE for scalable learning. Both explore relational data.
Evaluation Results
GraphSAGE: F1 Score 0.5213, Precision 0.6000, Recall 0.4609, ROC AUC 0.7141. Class imbalance noted.
Conclusion & Next Steps
GNNs show promise in churn prediction. Next: refine models, enhance visualisation.
Write down 3 things you learned in this lesson.
Write down 2 things you want to know more about.
Ask 1 question about something you haven't quite understood yet.