Case Study on CRM: Detecting Likely Churners with Limited Information of Fixed-line Subscriber

Abstract

Only with subscribers' contractual records and bill details in the investigated fixed-line services provider, we construct several churn prediction models, applying different data mining technologies. Comparing the models with predictors of duration of service use, payment type, amount and structure of monthly service fees and changes of the monthly service fees, we find that duration of service use is the most predictive variable. Then payment type and other predictors of amount and structure of monthly service fees, especially the predictors within the latest 3 months, are also effectual. We then build different decision tree models by reducing the amounts of monthly bill details for predicting, to find out how the prediction performance evolves. The result shows that with reduction of early monthly data, the model performance declines slightly. However, the processing data size and runtime of models decrease significantly. Thus, we suggest that using relatively fewer but latest monthly data to detect likely churner would be both effective and efficient

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 13 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.