![Page 1: Customer insights from telecom data using deep learning](https://reader035.vdocuments.us/reader035/viewer/2022062823/5878bc4c1a28ab724c8b7a67/html5/thumbnails/1.jpg)
Analytics on Telecom CDR Data
RedZebra AnalyticsOct 2014
![Page 2: Customer insights from telecom data using deep learning](https://reader035.vdocuments.us/reader035/viewer/2022062823/5878bc4c1a28ab724c8b7a67/html5/thumbnails/2.jpg)
Problem statement
1How to segment Telecom customers and track their dynamics
2How to optimize / reformulate tariff plans
3How to predict churn
![Page 3: Customer insights from telecom data using deep learning](https://reader035.vdocuments.us/reader035/viewer/2022062823/5878bc4c1a28ab724c8b7a67/html5/thumbnails/3.jpg)
The data
• 3 months of CDR– Data consumption– Phone calls and Topups– SMS
• User description (geo, sociodemographics)
![Page 4: Customer insights from telecom data using deep learning](https://reader035.vdocuments.us/reader035/viewer/2022062823/5878bc4c1a28ab724c8b7a67/html5/thumbnails/4.jpg)
The techniques
Deep Neural Networks and Autoencoders (Keras framework)
Random Forest
Extreme Gradient Boosting
Graph analysis (Igraph)
SOM and tSNE
Scikit Learn (Python)
![Page 5: Customer insights from telecom data using deep learning](https://reader035.vdocuments.us/reader035/viewer/2022062823/5878bc4c1a28ab724c8b7a67/html5/thumbnails/5.jpg)
Data processing (for churn prediction)
Churn (1) / no churn (0)
Customer activity is Converted into heatmaps
![Page 6: Customer insights from telecom data using deep learning](https://reader035.vdocuments.us/reader035/viewer/2022062823/5878bc4c1a28ab724c8b7a67/html5/thumbnails/6.jpg)
Network data also considered
We also include network data (like the number of churners connected to a node)
![Page 7: Customer insights from telecom data using deep learning](https://reader035.vdocuments.us/reader035/viewer/2022062823/5878bc4c1a28ab724c8b7a67/html5/thumbnails/7.jpg)
Three distinct users activity
![Page 8: Customer insights from telecom data using deep learning](https://reader035.vdocuments.us/reader035/viewer/2022062823/5878bc4c1a28ab724c8b7a67/html5/thumbnails/8.jpg)
Approach: Convolutional Neural Network
INPUTUser activityheatmap
OUTPUTChurn / no churn
![Page 9: Customer insights from telecom data using deep learning](https://reader035.vdocuments.us/reader035/viewer/2022062823/5878bc4c1a28ab724c8b7a67/html5/thumbnails/9.jpg)
Results
Method AUC - train AUC - testRandom Forest 0.75 0.74Extreme Gradient Boosting 0.80 0.76Variational Autoencoders 0.78 0.75Convolutional Neural Networks 0.79 0.77
Convolutional Neural Networks have the best performance
![Page 10: Customer insights from telecom data using deep learning](https://reader035.vdocuments.us/reader035/viewer/2022062823/5878bc4c1a28ab724c8b7a67/html5/thumbnails/10.jpg)
Some templates of user activity discovered by the neural network
![Page 11: Customer insights from telecom data using deep learning](https://reader035.vdocuments.us/reader035/viewer/2022062823/5878bc4c1a28ab724c8b7a67/html5/thumbnails/11.jpg)
SMS activity per age group
![Page 12: Customer insights from telecom data using deep learning](https://reader035.vdocuments.us/reader035/viewer/2022062823/5878bc4c1a28ab724c8b7a67/html5/thumbnails/12.jpg)
Clustering
Techniques used cluster and visualize data:• K-means• Self-organized maps (SOM)• tSNE
![Page 13: Customer insights from telecom data using deep learning](https://reader035.vdocuments.us/reader035/viewer/2022062823/5878bc4c1a28ab724c8b7a67/html5/thumbnails/13.jpg)
Visualization of sample of users with tSNE
![Page 14: Customer insights from telecom data using deep learning](https://reader035.vdocuments.us/reader035/viewer/2022062823/5878bc4c1a28ab724c8b7a67/html5/thumbnails/14.jpg)
Segmentation with Self Organized Maps
![Page 15: Customer insights from telecom data using deep learning](https://reader035.vdocuments.us/reader035/viewer/2022062823/5878bc4c1a28ab724c8b7a67/html5/thumbnails/15.jpg)
Distance to code-vectors: how stable is the population
![Page 16: Customer insights from telecom data using deep learning](https://reader035.vdocuments.us/reader035/viewer/2022062823/5878bc4c1a28ab724c8b7a67/html5/thumbnails/16.jpg)
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