Machine Learning in Telco - HFC Networks Predictive Maintenance
Download as PPTX, PDF
0 likes277 views
Predictive Maintenance of CMTS devices in HFC networks using Machine Learning and Random Forest algorithm.
Machine Learning in Telco - HFC Networks Predictive Maintenance
- 1. Machine Learning in Telco Belgrade, November 2017 Branko Kovac Senior R Developer & ML Engineer
- 2. Agenda Ibis Performance Insights Predictive Maintenance R Evangelism
- 3. Software Engineer -> Data Guy Data Scientist working at Ibis Instruments Co Founder of Data Science Serbia BelgradeR organiser About me
- 4. TelcoCATV Energy Enterprise Oil & GasSmart manufacturing Smart Cities Agriculture Predictive analytics / Machine Learning E2E Performance Monitoring SLA Packet Assurance Field Measurement Management Customer Analytics Ibis Performance Insights
- 6. Field Measurement Management Single point of view for all measurements results Technician efficiency monitoring Instrumentation utilization & performance report Field procedures monitoring
- 7. Predictive Analytics & ML Customer churn Cross-sell & upsell analytics Predictive Maintenance
- 8. CMTS Fault Prediction Fault here? Bad! :(
- 9. CMTS Faults Monitoring different metrics for faults Each metric has its own threshold(s) Signal to Noise Ratio (SNR) Metric is a time series
- 10. but I dont know (almost) anything about time series
- 12. ifelse(snr > 21, no, yes)
- 14. Feature engineering Keep it simple, ***pid! Mean, max, amplitude, sd... Number of samples in different ranges More than 60 different features Collinearity exists, but that can be handled
- 15. Random Forest Not sure about ML algorithm? RF! Easy to understand & powerful Results: accuracy 89,61% ~6,5% false alarms
- 16. Where is R in all this?
- 18. data.table Easily handles ~100 million rows Fast, memory efficient dplyr is nice, but not a solution if dataset is large Thanks to Matt & Arun!
- 19. mlR Standardized interface for ML Concept of tasks & learners Rich preprocessing & customization Bayesian model optimization via mlrMBO Thanks to Bernd & others!
Editor's Notes
- #2: Thanks for coming Three years ago when conference was started
- #5: Vendor agnostic, scalable, on premise or in cloud IN Telco used in NOC & field tech guys
- #6: E2E perf monitoring tool for monitoring all devices in a network; intelligent & non-intelligent devices; we get data about non-intelligent by counting number of working intelligent devices connected to it Report about condition of a network, totally customizable Focus on a network segment; red not working, green working
- #7: Collect all information about tech guys working in a field Measure efficiency of tech people and provide reports Make decisions to improve quality of work and employee effiency
- #8: Who will cancel subscribtion Additional services for better experience; personalized offers High availability is important can we predict faults and improve services
- #9: Hybrid fiber coaxial network Internet, IPTV