Outline
- data types/concepts
- people behaviour
- tuning/predictions
Data privacy
- handover between cells - calls/internet package
- no content/size information
- new hashed id every day
- no group (age, device, zip...) below 30 people
event collection
We collect an handover between cells per device (D1-Netz)
Events to movements
We groups events into activities and trips
Activity probability
Density probability of activities between cells
User's behaviour
Different signaling patterns during the day
Daily patterns
People have different patterns every day
Tourist patterns
Split nationality, origin, counting activities
age, gender, device types
Distribution of age classes per zip code
From cells to routes
- Interpret cell connection during a trip
- Understand mode of transportation
- Calculate most probable route
- Count people
movements across the country
How people move during a day
Most frequent origin and destination
important for local transportation companies
routing/infrastructure
we work on an efficient routing and infrastructure
local network
Special modules consider city centers
subway module
Distinguish the mean of transportation labelling cells (training data)
subway passangers
Collecting information about commuters
Use of information
- activities, footfalls, section counts
- points of interest/geo context
- statistical geographical population data
- public/training reference data
Intensity of activities
Activity intensity: breaks of motorway drivers
geographical data
Validate results with statistical data (population density)
collect important features
we select the most relevant features
training set
geographical information to train system
training
We train our models to predict success
compare external sources
Public information from different sources
forecast
We use past time series for forecasting events
mapping on offer
Calculate the coverage of a commercial activity
whitespot analysis
Customer potential on regions not covered within a isochrone
Thank you
for your attention
