MOBILE PHONE ACTIVITY
MILAN, ITALY
Peyman Hesami
DSE241 Final Project
MOTIVATIONS
• Mobile phone activities generate massive amount of data
• This can be used in mobility planning, tourist flows, urban structures
and interactions, event detection, urban well-being and many others
MOTIVATIONS
• It can also be used for cellular network diagnostics and maintenance
• Finding congested cells/areas
• Finding idle cells/areas
• Finding user’s usage pattern
DATASET
• One week of Call Details Records (CDRs) from the city of Milan and the Province of Trentino (Italy)- 1.5
GB
• Both domestic (Milan to other provinces) and international (Milan to other countries) data
• Third source of data:
• (lat, long) of countries and provinces of Italy
• Geojson file of Milan cellular network containing (lat, long) of cells in the city of Milan
DATA WRANGLING
• Converting the raw dataset to two sets of nodes/edges dataset
• Adding label to the nodes based on their type (Milan, domestic, international)
• Removing edges with no 0 values (for sms, call, …)
• Extracting day and hour from date time
• Integrating third source of data:
Converting country codes to country names
Deriving the coordinates (lat and long) of the cells in Milan, provinces of Italy,
and other countries
TASKS
• Goals:
1. Diagnostic tool: Visualization tool to help
wireless network engineers in cellular networks
diagnostics
2. Presentation tool: User friendly
representation of the mobile phone activities for
nontechnical presentation
TASKS-DIAGNOSTIC TOOL
• A node-link diagram with:
• Nodes as cells and edges as user activities (width~magnitude channel)
• Several filters to choose the type of the data; domestic vs international, sms vs call vs Internet
• Zoom in/out capability
• Interaction with the graph by selecting nodes and highlighting their connected edge, adding labels to
nodes/edges
• Adding time sliders to choose the time interval within a day, day of the week and animation across time
• Ability to choose the desired graph layout
• Ability to show only the most significant edges based on a user input
• Ability to hide nodes and edges on drag for easy interactions
AUDIENCE-DIAGNOSTIC
TOOL
• Cellular network engineers:
• Identifying the troubled (congested) cells
• Identifying the idle (inactive) cells
Optimize them to serve more users
• Identifying the cell use pattern across days of the weeks and hours of the day
Schedule maintenance time in low traffic time intervals.
• Identifying the data usage patterns across time and geography
Optimize the cells dynamically based on usage
TASKS-PRESENTATION TOOL
• Great Circle on geo layout with:
• points as cells and lines as user activities
• Several filters to choose the type of the data; domestic vs international, sms vs call vs Internet
• Adding time sliders to choose the time interval within a day, day of the week and animation across time
• Ability to show only the most significant edges based on a user input
AUDIENCE-PRESENTATION
TOOL
• Nontechnical users seeking:
• To find mobile phone user’s usage patterns across time and
geography.
• Study the usage pattern alongside other sources of the data (like
census data) for socio-technical analysis (like targeted marketing)
SOLUTIONS
• Visualization type
• Data reduction (filtering)
• Data reduction (sampling)
SOLUTIONS
• Data Reduction (edge filtering)
• View change over time (static and animation)
SOLUTIONS
• Data reduction (aggregation)
• Graph layout
• Graph annotations/features
• Data reduction (edge filtering)
IMPLEMENTATION
• Shiny package in R
• Run ui.R or server.R in RStudio
• Deployed on shiny.io serve: https://peymanshiny.shinyapps.io/
milan_phone_activity_shiny_dse241_peyman_hesami/
RESULTS
RESULTS
• Comparing network at 1am and 3pm on a Monday
RESULTS
• Comparing outgoing calls on weekdays versus weekend at 3pm
CHALLENGES AND
IMPROVEMENTS
• Efficient reading data into memory is required for fast
user interactions (multiple libraries tried)
• Hour/Day extraction can be costly (regex)
• Great circle vis is not completely interactive
• Other sources of data (census) can be integrated for
more insights
