Mais conteúdo relacionado Semelhante a Predictive Analytics in Telecommunication (20) Mais de Rising Media Ltd. (20) Predictive Analytics in Telecommunication1. 07/11/20141 © Nokia 2014 T&I Research / Norbert Kraft
Predictive Analytics in
Telecommunication
Norbert Kraft
Nokia Technology & Innovation
2. 07/11/20142 © Nokia 2014 T&I Research / Norbert Kraft
Short Introduction
• Researcher at Nokia Technology & Innovation
• Project Leader
‘Network Data Intelligence’
3. 07/11/20143 © Nokia 2014 T&I Research / Norbert Kraft
Network Data Intelligence
• Nokia Research Project
• Technology exploration
• Generate new insights in telecom data
• Raise new business opportunities
Mobile
Networks
Data
Mining
Machine
Learning
Big Data
4. 07/11/20144 © Nokia 2014 T&I Research / Norbert Kraft
Why: Requirements in Telecommunication
What: Use Cases
How: Ways to get it done
Problems and Outlook
Agenda
Predictive Analytics in Telecommunication
5. 07/11/20145 © Nokia 2014 T&I Research / Norbert Kraft
THREE BUSINESSES AT THE HEART OF
THE COMING CONNECTED WORLD
NOKIA
NETWORKS
End-to-end mobile
broadband and services
• Programmability
• Hardware to
software
• Big data analytics
• Virtualization and
cloud
Advanced R&D and IP for
licensing and new product
businesses
• Enabling new customer
experiences
• Sensing and materials
• Connectivity and actuation
Making the map of the future
the source of location
intelligence
• Map content assets
• Location platform
• Relevant, seamless
user experiences
NOKIA
TECHNOLOGIESHERE
6. 07/11/20146 © Nokia 2014 T&I Research / Norbert Kraft
… More Than an End Device
End to End Mobile Broadband
Dock + O/E
conversion
Mini BTS
Standalone GPS
module
External
directional
antenna
User Entity BTS/eNodeB P-GWS-GW
MME HSS PCRF
Internet
User Data
Signaling
7. 07/11/20147 © Nokia 2014 T&I Research / Norbert Kraft
• 36.6 Million Subscribers for German
Telekom
• Total of 113 Million Subscribers in Germany
• 70 000 Radio Cells in Germany
• 100 Million GBytes traffic volume (*2011)
• xxx.xxx.xxx.xxx Number of Calls & SMS per
Day
• xxx.xxx.xxx.xxx Number of Internet
connections
• SmartPhone is always ‘ON’
Some (estimated) Numbers …
German Telekom (2012)
Source: Bundesnetzagentur from 2012
Radio Cell Layout of Munich
8. 07/11/20148 © Nokia 2014 T&I Research / Norbert Kraft
Total number of Radio Cells: Munich Example
9. 07/11/20149 © Nokia 2014 T&I Research / Norbert Kraft
Why: Requirements in Telecommuniation
What: Use Cases
How: Ways to get it done
Problems and Outlook
Agenda
Predictive Analytics in Telecommunication
10. 07/11/201410 © Nokia 2014 T&I Research / Norbert Kraft
Reasons to Talk about … in Telco Space
Predictive Near Real Time Big Data Analytics
Predictive
•From reactive to pro-active mode
•Don’t detect - avoid problems
Near Real Time
Big Data Analytics
•Support calling customer at once
•Most use cases have real time
aspect
•XX.XXX.XXX subscribers
•XX.XXX radio cells
•Any service affects several systems
•Modern users are always on
•Solve ‘The 5000 KPI’ problem
•Detect hidden problems
•Find root causes of problems
•A single problem causes xxx alarms
11. 07/11/201411 © Nokia 2014 T&I Research / Norbert Kraft
User
Mobile Network Data on ‘Signaling’
What the Operator (needs to …) know about …
User
Identity
Location
Service
Usage
Data
Volumes
Network
Personal
Data
Network
Element
Status
Configuration
Data
Performance
values
Alarms
SW Logs
& Traces
CDRs
IMSI
Device ID &
Type IMEI
MSIDN
Phone
NOs A/B
Cell location
XXX m
Higher precision w.
triangulation on
signal strength
URL
User
Agent
IP / port
addresses
Tarif
Address
Revenue
Call/SMS
Length
Bytes
up/down
load
structured unstructured
Highly
structured
12. 07/11/201412 © Nokia 2014 T&I Research / Norbert Kraft
Network Data is Personal
Data
Disclaimer
> Strictly limited by (inter)national laws
> Very complex field under continuous change
> Different views in different countries
> Restrictions on use beyond network
management scope
> Usage requires customer permission
> Network operators have the right to use this
data for management purposes
> Billing
> Fault diagnosis
> Network improvement
> Support activities
!!!! But
13. 07/11/201413 © Nokia 2014 T&I Research / Norbert Kraft
Map of Big Data Analytics Use Cases
Network
Planning
Radio cell
Performance
User Mobility
WiFi offload
Drop Call
Probability
High-volume
applications
High-volume
websites
Peak data
information
Roaming
analysis
Operation
Failure
analysis
Predict
network
outages
Video
download
experience
Service
failures
Predictive HW
maintenance
Chronic circuit
problems
Security
BotNet
detection
Intrusion
detection
DOS attacks
Customer
Product
management
OTT tracking
Tarif simulation
Verifying new
services, products &
devices
Viral marketing
CRM
Fraud
detection
Churn
probability
Customer
Segmentation
Loyalty offers
Service Up
selling
Tracking specific
customers (VIPs,
dissatisfied)
Service
First best offer
Feedback
analysis
Discourage
SIM swapping
Pre-pay
recharge
message
Personalized
portal
Troubleshootin
g support
Bill shock
messages
External
Public
interest
Disaster
detection
Crowd
estimation
Traffic
supervision
Big event
management
Commercial
interest
Advertisement
User mobility
transportation, traffic
planning, ads
Site
protection/sup
ervision
Large site
management
Off topic
Social media
Sentiment
analysis
Peer group
analysis
Picture recognition
information
extraction
14. 07/11/201414 © Nokia 2014 T&I Research / Norbert Kraft
• Important SLA
criteria
• History Needs to
be continuously
monitored
• Prediction turns
monitoring to
preventive activity
Network Use Cases
Dropped Packet Connections per Radio Cell
15. 07/11/201415 © Nokia 2014 T&I Research / Norbert Kraft
• Movement vectors
- Color:
• Direction
• speed
- Thickness: no of users
• Usage
- Traffic planning
- Ads planning
- Traffic jam prediction
Network Use Cases
User Mobility
16. 07/11/201416 © Nokia 2014 T&I Research / Norbert Kraft
• Cell classification
- Business
- Private home area
• Predicts preferred
user movement
• Usage
- WiFi offload
Network Use Cases
User Mobility & WiFi Offload
17. 07/11/201417 © Nokia 2014 T&I Research / Norbert Kraft
• Time slot
classification on
history
- Normal behavior
- Outliers
• Long term trend
analysis
• KPI radar &
prediction
Network Use Cases
KPI Prediction & Time Slot Classification
18. 07/11/201418 © Nokia 2014 T&I Research / Norbert Kraft
• Show as many
dimensions as
possible
• Show relations
between data
Network Use Cases
Parameter Correlation
19. 07/11/201419 © Nokia 2014 T&I Research / Norbert Kraft
• Service KPIs monitor
system health
• SLA agreement
guarantees 99.x %
availability
• Steady values for
normal operation
Network Use Cases
Predictive Operations
20. 07/11/201420 © Nokia 2014 T&I Research / Norbert Kraft
• Early warning
indicators
• From fault detection to
fault avoidance
• Methods:
- E.g. Hidden Markov
Chains
• Accuracy: ~80-85%
Network Use Cases
Predictive Operations
KPI
Drop
Anom
alies
21. 07/11/201421 © Nokia 2014 T&I Research / Norbert Kraft
• 1000 KPIs don’t
help ….
• Root cause
- Driving value ?
- Origin ?
- Combinations of
KPIs
• Techniques
- Decision Tree
• Used for prediction
of service failures
Network Use Cases
Root Cause Analysis
Network Use Cases
22. 07/11/201422 © Nokia 2014 T&I Research / Norbert Kraft
• An outage always impacts a lot of customers
• Predict hardware failures in advance
• Collect triggers (possibly) signaling an outage:
- SW logs/traces
- Hardware signals/deviations (Temperature …)
• Find (performance/behavior) outliers in the
whole set of all net elements
• Replace in advance
Network Use Cases
Predictive Hardware Maintenance
Dock + O/E
conversion
Mini BTS
Standalone GPS
module
External
directional
antenna
23. 07/11/201423 © Nokia 2014 T&I Research / Norbert Kraft
Why: Requirements in Telecommuniation
What: Use Cases
How: Ways to get it done
Problems and Outlook
Agenda
Predictive Analytics in Telecommunication
24. 07/11/201424 © Nokia 2014 T&I Research / Norbert Kraft
PredictiveBig
DataAnalytics
Stack
Generalized Data Analytics Stack
Important Components
Import, Formatting, Type Conversion
Aggregation, Filtering, Distributed Computing
Analytics Algorithms (K-Means, KNN…)
Visualization, Charting, Drill Down Views
Use Cases
Data Storage (Relational, NoSQL)
Data Sources: net elements, protocols
25. 07/11/201425 © Nokia 2014 T&I Research / Norbert Kraft
PredictiveBig
DataAnalytics
Stack
Generalized Data Analytics Stack
NDI Components
Import, Formatting, Type Conversion
Aggregation, Filtering, Distributed Computing
Analytics Algorithms (K-Means, KNN…)
Visualization, Charting, Drill Down Views
Use Cases
Data Storage (Relational, NoSQL)
Data Sources: net elements, protocols
Development Effort
26. 07/11/201426 © Nokia 2014 T&I Research / Norbert Kraft
PredictiveBig
DataAnalytics
Stack
Generalized Data Analytics Stack
Realization Components
Import, Formatting, Type Conversion
Aggregation, Filtering, Distributed
Computing
Analytics Algorithms (K-Means,
KNN…)
Visualization, Charting, Drill Down
Views
Use Cases
Data Storage (Relational, NoSQL)
Data Sources: net elements, protocols
Analytic
Tools
RapidMine
r
Knime
SPSS
Parallel
processing
Hadoop,
Storm
Hive
Pig
Tableau,
QlikView
Mahout
Language
Stacks
R
Python
SCI-Kit
(Python)
Pandas
(Python)
Big Data
DBs
Oracle
Teradata
NoSQLs
27. 07/11/201427 © Nokia 2014 T&I Research / Norbert Kraft
PredictiveBig
DataAnalytics
Stack
Generalized Data Analytics Stack
Realization Components
Import, Formatting, Type Conversion
Aggregation, Filtering, Distributed
Computing
Analytics Algorithms (K-Means,
KNN…)
Visualization, Charting, Drill Down
Views
Use Cases
Data Storage (Relational, NoSQL)
Data Sources: net elements, protocols
Analytic
Tools
RapidMine
r
Knime
SPSS
Parallel
processing
Hadoop,
Storm
Hive
Pig
Tableau,
QlikView
Mahout
Language
Stacks
R
Python
SCI-Kit
(Python)
Pandas
(Python)
Big Data
DBs
Oracle
Teradata
NoSQLs
Vertica
Needs to be
complemented
by powerful DB
Problems with
Big Data
High entry
barrier
No DB
replacement
Requires
good analyst
background
Not very
popular
Crowded
place
28. 07/11/201428 © Nokia 2014 T&I Research / Norbert Kraft
Data Storage Approaches
Which Store to take for Big Data Analytics
Relational
• Stable, well
known
• Great features
• Optimized for
lots of parallel
transactions
• Horizontal
scaling
Key/value store
• Fast read /
write
• Missing
aggregation
capabilities
• No multi
indexing
Column oriented
• Made for fast
aggregation
• Vertical scaling
Document
oriented
• No fixed
schema
• Alternative
aggregation
engines
New kids on the
block / Hadoop
• Missing lots of
db features
• High
throughput
processing
• No built-in
aggregation
functions
• No real time
support
• Huge eco
system
29. 07/11/201429 © Nokia 2014 T&I Research / Norbert Kraft
DataAnalyticsStack
Generalized Data Analytics Stack
NDI Components
Import, Formatting, Type Conversion
Aggregation, Filtering, Distr.Comp.
Analytics Algorithms (K-Means, KNN…)
Visualization, Charting, Drill Down
Use Cases
Data Storage (Relational, NoSQL)
NDI – Distributed Real Time Importer
Data Sources: net elements, protocols
Service
KPIs
DPI
Data
Enrichments
OpenCellId, TAC, Locations
Network Element Data
NDI – Server
NDI – Client
Analytics
Engine
User
Mobility
BS
Synchronization
Root Cause
Analysis
Predictive
Operation
SQL DBs
NoSQL
DBs
Service
Dashboard
30. 07/11/201430 © Nokia 2014 T&I Research / Norbert Kraft
Development
Environment
Real Time
Streaming
Server
Network Data Intelligence Demonstrator
Architecture
Database Layer
MySQL Others
Django
Python
Browser Client
JavaScript
OpenLayers
HTML
CSS
HighCharts
Pandas
RapidMiner
Mongo
DB
HTTP
REST
Rapid
Miner
Server
Tableau
Desktop
OrangeTouch
JQuery
D3
RAW
Data
Artificial Data
Generator
SCI Kit
Map
Reduce SQL
Python
PandasSCI Kit
Standard Programming
Data Analytics & Aggregation
Rich Client & Charting
Tool
32. 07/11/201432 © Nokia 2014 T&I Research / Norbert Kraft
Confidential
NDI Parallel Real Time Engine
Scaling Architecture
DPI
Net
Element
Net
Element
KPIs
Enrich
ments
N * Real Time
Engine workers
N * Real Time
Engine workers
N * Real Time
Engine workers
N * Real Time
Engine workers
N * Real Time
Engine workers
NDI Server
N * Real Time
Engine feeders
N * Real Time
Engine feeders
Message
Broker
NDI Rich
Client
NDI Real Time Importer
33. 07/11/201433 © Nokia 2014 T&I Research / Norbert Kraft
Why: Requirements in Telecommuniation
What: Use Cases
How: Ways to get it done
Problems and Outlook
Agenda
Predictive Analytics in Telecommunication
34. 07/11/201434 © Nokia 2014 T&I Research / Norbert Kraft
Things not solved so far …
Challenges Volume
Velocity
Variety
Veracity
Data Size
• Massive parallel processing
Data Formats & Sources
• ???
Data in Motion
• Stream processing
4Vs
In Big
Data
Data in Doubt
• Cleaning, filtering
• ???
4Vs
in
Big Data
Analytics
36. 07/11/201436 © Nokia 2014 T&I Research / Norbert Kraft
… And Keep in Mind
Predictive Analytics is Only the First Step!
Repair
37. 07/11/201437 © Nokia 2014 T&I Research / Norbert Kraft
Thank You!
Questions?
norbert.kraft@nsn.com