Coefficient of Thermal Expansion and their Importance.pptx
Master thesis
1. A Visual Analytics Framework for
Anomaly Analysis and Prediction in Mobile
Telecommunication Scenario
Candidate
Matteo Stabile
ID Number: 1547019
Co-Advisor
Marco Angelini
Advisor
Giuseppe Santucci
Academic Year: 2016/2017
2. 30/01/2018Matteo Stabile Pagina 2
Introduction: problem and approach
Telecom Italia dataset
Visual encoding
Demo
Evaluation
Conclusions and future works
Outline
3. 30/01/2018Matteo Stabile Pagina 3
Introduction
Problem: handling
hundreds of thousands
of phone cells for
monitoring, analysis
and impact prediction
support.
Monitoring and analysis of network
related aspects:
➢ around 230000 phone cells;
➢ thousands of failures alarms per-day.
Impact prediction support: stakeholder
required a solution to support failures
fixing strategies.
«Which failures should be fixed first?»
Problem
4. 30/01/2018Matteo Stabile Pagina 4
Introduction
Approach: visual
analytics framework
based on two main
guidelines.
Explorative environment analysis:
reducing the total amount of objects
from 230000 to at most 1600 !!!
Impact prediction support model:
it is based on statistical data
concerning average behavior of
each single cell during a week day.
Province of Rome:
• Before: 2855 points
• Now: 4 squares
Province of Milan:
• Before: 3970 points
• Now: 4 squares
32x32 square sizes
Approach
5. 30/01/2018Matteo Stabile Pagina 5
Introduction: problem and approach
Telecom Italia dataset
Visual encoding
Demo
Evaluation
Conclusions and future works
Outline
6. 30/01/2018Matteo Stabile Pagina 6
Dataset description
➢ Confidential dataset by Telecom Italia.
➢ Multidimensional information:
Cells
▪ Total amount: 229038
▪ Distributed as expected
▪ Total features: 35
Presences ▪ Number of cells connections
▪ Total features: 8
▪ Grouped by 15-minutes intervals
Failures
▪ 8800 failures per-day
▪ Total features: 13
▪ Grouped by 15-minutes intervals
7. 30/01/2018Matteo Stabile Pagina 7
▪ Introduction: problem and approach
▪ Telecom Italia dataset
▪ Visual encoding
Demo
Evaluation
Conclusions and future works
Outline
8. 30/01/2018Matteo Stabile Pagina 8
Visual encoding: system overview
Cells representation view (map-based)
Failures representation view (gantt-based)
10. 30/01/2018Matteo Stabile Pagina 10
Visual encoding: map-based environment
Italy map: grid sectioning
representation.
Statistical tools: boxplot
and color legend for
inspecting the same
underlying population.
Categorical filters: for choosing the kind of
information on foreground map.
Squares size options:
- «Auto»: the system
chooses the best size
for the current map
zoom level.
- Users manually
chooses the interesting
sizes, despite the map
zoom level.
11. 30/01/2018Matteo Stabile Pagina 11
Visual encoding: gantt-based environment
Current mode:
• Historical analysis:
analysis of previous
network status.
• Monitoring: analysis of
future system condition.
Time lines:
• Yellow line for representing past time intervals.
• White line for representing future timestamps.
Time
Priority
Cell name Next peak of
presences
Opening time
Time left
Time elapsed
Older peak of
presences
12. 30/01/2018Matteo Stabile Pagina 12
Priority criteria
Visual encoding: gantt-based environment
Time left based criterion (ΔT):
gives priority to the failure with
the sorthest time left. Presences based criterion (P):
gives priority to the failure with
highest amount of affected
presences.
13. 30/01/2018Matteo Stabile Pagina 13
▪ Introduction: problem and approach
▪ Telecom Italia dataset
▪ Visual encoding
Demo
Evaluation
Conclusions and future works
Outline
15. 30/01/2018Matteo Stabile Pagina 15
▪ Introduction: problem and approach
▪ Telecom Italia dataset
▪ Visual encoding
Demo
Evaluation
Conclusions and future works
Outline
16. 30/01/2018Matteo Stabile Pagina 16
Evaluation
➢ Agile-based context:
▪ Total duration: 7 months
▪ Two main milestones
▪ Second milestone outcomes:
First at month 4
Second at month 6
Requirements
The need to have a more robust
views coordination.
Better representation of temporal
concepts (historical analysis and
simulation).
The need to provide data filtering
mechanisms for operators.
Proposed solutions
Mouse events trigger data filtering
process, in both the views directions.
Two time lines define past time
stamps and «future» ones.
View settings panel allows to decide
displayed data categories.
17. 30/01/2018Matteo Stabile Pagina 17
▪ Introduction: problem and approach
▪ Telecom Italia dataset
▪ Visual encoding
Demo
Evaluation
Conclusions and future works
Outline
18. 30/01/2018Matteo Stabile Pagina 18
Conclusions and future works
Impact predictive
framework based on visual
analytics techniques.
Background-awareness
application context.
It is interesting to improve the
predictability spectrum by historical
analyzing past fixing activities.
Studying how is possible to use
statistical filters in conjunction.
Future worksConclusions