relating mobility patterns to socio demographic profiles
DESCRIPTION
Thomas Liebig Thomas Liebig/Technical University of Dortmund, Germany. Topic: “Relating mobility patterns with socio-demographic profiles”TRANSCRIPT
Relating mobility patterns to socio-demographic profiles
Thomas Liebig
Artificial Intelligence Group
Technical University of Dortmund
Agent Based Simulation
Models individual mobility with artificial agents
Hierarchy of motion [Hoogendoorn et.al. 2002]
Agent Based Simulation
Models individual mobility with artificial agents
Required data
‣ Traffic Network Traffic Network
‣ Points of interest Facilities
‣ Description of population Plans
Models individual mobility
Required data
‣ Traffic Network
‣ Points of interest
‣ Description of population
Agent Based Simulation
Description of population
‣ Denotes individual plan for every agent
‣ Where do we get it from?
‣ We present a method to derive it from census data
and analysis of sample dataset (e.g. CDC2013)
Different people in town
‣ Who prefers which mobility behaviour?
Analysis Workflow
‣ Process recorded raw data to sequences of annotated
locations
‣ filtering, stop detection, clustering, labeling
‣ Identify most frequent sequences and their
supporting subgroup of the population
Input data:
PID, Sequence (home, work, home)|demographic attributes (gender, age, …)
Frequent Itemset Mining
‣ We apply FP-tree
requires threshold (in this example threshold=1)
‣ {1, 3, 4}
{2, 4, 5}
{2, 4, 6}
‣ 6,26,246,…
Subgroup Analysis
‣ technique for the extraction of patterns
‣ with respect to a target variable.
‣ describes relations between variables and a certain value
of the target variable.
Frequent pattern {2,4}
‣ {1, 3, 4,x=0} male, student
{2, 4, 5,x=1} female, worker
{2, 4, 6,x=1} male, worker pattern: worker,{2,4}
Subgroup Analysis
‣ technique for the extraction of patterns
‣ with respect to a target variable.
‣ describes relations between variables and a certain value
of the target variable.
Frequent pattern {2,4}
‣ {1, 3, 4,x=0} male, student
{2, 4, 5,x=1} female, worker
{2, 4, 6,x=1} male, worker pattern: worker,{2,4}
Test with cyclists data
‣ given are trips with their purpose and person identifier
‣ About 80 persons
‣ purposes To work
To visit (friends, etc);
To work related task;
To Food shopping;
To Non-food shopping;
To School (Student);
To Entertainment;
To Eat (Lunch, etc);
To Home;
Other (any other not mentioned)]
‣ For the persons several attributes are provided gender, age, health, employment, income, marital status
(changed to binomial attributes)
Result - Freqent patterns
‣ Threshold: 0.25
‣ 69 To work and to home
37 To home and to Other
30 To work and to Other
30 To work and to home and to other
29 To Home and To Eat
23 To Work and to Eat and to Home
…
Result - Subgroups
‣ 69 To work and to home
TRUE, for
27-30 years=false and
FullTimeEducation=false and
Employment_Other=false and
SelfEmployed=false and
Income Low=false
37 To home and to Other
30 To work and to Other
30 To work and to home and to other
29 To Home and To Eat
23 To Work and to Eat and to Home
…
Result - Subgroups
‣ 69 To work and to home
37 To home and to Other
FALSE for
23-26 years=false and
35-38 years=false and
55-58 years=false and
EmployedPartTime=false and
Employment Other=false
30 To work and to Other
30 To work and to home and to other
29 To Home and To Eat
23 To Work and to Eat and to Home
…
Result - Subgroups
‣ 69 To work and to home
37 To home and to Other
30 To work and to Other
TRUE, for
27-30 years=false and
47-50 years=false and
51-54 years=false and
SelfEmployed=false and
Income Medium=false
30 To work and to home and to other
29 To Home and To Eat
23 To Work and to Eat and to Home
…
Result - Subgroups
‣ 69 To work and to home
37 To home and to Other
30 To work and to Other
30 To work and to home and to other
TRUE, for
27-30 years=false and
47-50 years=false and
51-54 years=false and
SelfEmployed=false and
Income Medium=false
29 To Home and To Eat
23 To Work and to Eat and to Home
…
Result - Subgroups
‣ 69 To work and to home
37 To home and to Other
30 To work and to Other
30 To work and to home and to other
29 To Home and To Eat
FALSE, for
Single=false and
Health Fair=false and
Employment Other=false and
SelfEmployed=false and
Income Low=false
23 To Work and to Eat and to Home
…
Result - Subgroups
‣ 69 To work and to home
37 To home and to Other
30 To work and to Other
30 To work and to home and to other
29 To Home and To Eat
23 To Work and to Eat and to Home
TRUE for
43-46 years=false and
Married=false and
Health-Very Good=false and
FullTimeEducation=false
…
Summary
‣ Found patterns can be used to define plans for agents
based on census of a city
(e.g. for mode of transportation decisions)
‣ Application to CDC2013
Next steps
‣ Spatio-Temporal Subgroups
‣ Performance analysis