Explainable Multi-Stage Churn Prediction using Graph Neural Network in Telecom Sector

Customer churn prediction is a key task for telecom companies, as retaining customer is more effective than getting new customers. Traditional churn prediction model often lacks the ability to provide insights, making it hard to understand the factors leading for customer churn. This project proposes an explainable multi-stage churn prediction model using Graph Neural Networks (GNNs) to enhance both accuracy and efficiency of predictions made in the telecom industry. This approach models customer interactions as graph, capturing complex relations and dependencies between customers. Using techniques like LIME and SHAP, the model provides clear, actionable explanations for decision making. Experimental results demonstrate GNN based model maintained high level of explainability, making it an effective tool for telecom companies to manage churn and improve customer retain strategies.