Customer insights from telecom data using deep learning
Download as PPTX, PDF3 likes1,598 views
This document discusses using analytics on telecom call detail record (CDR) data to segment customers, optimize tariff plans, and predict churn. It analyzes 3 months of CDR data including data usage, calls, SMS, and user demographics using techniques like deep neural networks, random forest, and graph analysis. Customer activity heatmaps are converted to churn prediction inputs for a convolutional neural network model. The techniques revealed user activity templates and clustered customers into stable segments to help telecom companies optimize plans and reduce churn.
Customer insights from telecom data using deep learning
- 1. Analytics on Telecom CDR Data RedZebra Analytics Oct 2014
- 2. Problem statement 1How to segment Telecom customers and track their dynamics 1How to optimize / reformulate tariff plans 1How to predict churn
- 3. The data •3 months of CDR –Data consumption –Phone calls and Topups –SMS •User description (geo, sociodemographics)
- 4. The techniques Deep Neural Networks and Autoencoders (Keras framework) Random Forest Extreme Gradient Boosting Graph analysis (Igraph) SOM and tSNE Scikit Learn (Python)
- 5. Data processing (for churn prediction) Churn (1) / no churn (0) Customer activity is Converted into heatmaps
- 6. Network data also considered We also include network data (like the number of churners connected to a node)
- 7. Three distinct users activity
- 8. Approach: Convolutional Neural Network INPUT User activity heatmap OUTPUT Churn / no churn
- 9. Results Method AUC - train AUC - test Random Forest 0.75 0.74 Extreme Gradient Boosting 0.80 0.76 Variational Autoencoders 0.78 0.75 Convolutional Neural Networks 0.79 0.77 Convolutional Neural Networks have the best performance
- 10. Some templates of user activity discovered by the neural network
- 11. SMS activity per age group
- 12. Clustering Techniques used cluster and visualize data: •K-means •Self-organized maps (SOM) •tSNE
- 13. Visualization of sample of users with tSNE
- 14. Segmentation with Self Organized Maps
- 15. Distance to code-vectors: how stable is the population
- 16. Conclusions •Deep Convolutional Networks achieve top performance •Network data very important (who is connected to who) •We found 5 well defined segments •Payments are determined by calls not data •SOM create relatively stable segments •Intercommunity diverse is some cases