A New Generalized Mixed Data Model with Applications to Transport Analysis

A New Generalized Mixed Data Model with

Applications to Transport Analysis

Chandra Bhat

Research partially supported by

• The U.S. DOT through the Data-Supported Transportation Operations and Planning (D-STOP) Tier 1 Center

• Alexander von Humboldt Foundation, Germany

Introduction and Motivation

• Growing interest in joint modeling of data with mixed types of

dependent variables in several fields

• Clinical biology: effectiveness of depression medication in

reducing occurrence, frequency, and intensity of depression

• Health: occurrence, frequency, and intensity of specific health

problems, as well as ordinal quality of life

• Transportation: Translating voluminous amounts of data into

information in near-real time or for planning purposes to take

proactive action

Data Science

• Not enough humans to process

• Machine learning, visualization, and advanced computation

techniques

• Statistics, social sciences, and domain knowledge

Why joint modeling is important?

• Borrows information on other outcomes

• Able to answer intrinsically multivariate questions, such as the

effect of a covariate on a multidimensional outcome

• Is able to integrate data to increase accuracy as well as

precision of information extraction.

• Helps causal effects to be distinguished from associative

effects.

• The new Generalized Heterogeneous Data Model (GHDM).

• Correlation across various dimensions are captured using latentconstructs.

• Accommodates all types of data (independent and dependentvariables).

• Bhat (2014) on Composite Marginal Likelihood (MACML)

• High dimensional independent variable setting (operations)

• High dimensional dependent variable setting (planning)

Connected vehicles technology provides high dimensional heterogeneous data

Vehicles have embedded Computers and GPS receivers short-range wireless network interfaces in-car sensors, cameras, and internet

Vehicles interact with Roadside wireless sensor networks other cars Other road-users.

Localized versus Central Data Processing and Analysis

Methodologies to translate data into information

COLLABORATE. INNOVATE. EDUCATE.

Data required to keep vehicle safely on the road

Highly detailed maps information: Shape and elevation of roadways, lane lines, intersections, crosswalks, speed limits, and traffic signals.

Position, speed and intentions of other vehicles and pedestrians.

Position, speed and intentions of unexpected obstacles, such as, jaywalking pedestrians, cars lunching out of hidden driveways, a stop sign held up by a crossing guard, and cyclist making gestures.

A simple example (operations)

• Assume two vehicles and an isolated non-signalized intersection

• Assume all measurements captured precisely

Position of Vehicle 1

(binary/continuous)

Speed of Vehicle 2

(continuous)Position of Vehicle 2

(binary/continuous)

Speed of Vehicle 1

(continuous)

Direction

and angle of

progress of

Veh. 1

Direction and

angle of

progress of

Veh. 2

Vehicle 1

type/Age

(nominal, binary)

Vehicle 2

type/age

(nominal,

binary)

Weather

conditions

Convergence

rate index

Vehicle

separability index

Crash Occurrence

(yes/no)

• Position/trajectories of other vehicles

• Human in the loop

• Probability model (multi-index decision variable modeling)

• Projection: Principal components of a covariance matrixconstructed from the sub-samples of crashes and no crashes

• Estimation: Parametric or non-parametric choice modeling

COLLABORATE. INNOVATE. EDUCATE.

Lane-departure detection

Mechanism to detect when another vehicle begins to move out of its lane.

Minimize accidents by addressing the main cause of collisions, driving errors, and distractions.

COLLABORATE. INNOVATE. EDUCATE.

Automatic braking

Sensor to detect an imminent collision with another vehicle, person or obstacle.

Car actives the brakes itself.

COLLABORATE. INNOVATE. EDUCATE.

Self-parking

Car parks itself.

Drivers do not need to worry about finding a parking spot.

A simple example (planning)

• Consider residential choice and activity-travel behavior today

• Expansion in focus: Proactive, demand reducing, short-term,

sustainability-oriented

• Emphasis on land-use and transportation

policies to shape travel behavior

• Over the past decade

• Increasing attention on the causal vs.

associative nature of the relationship

• Residential self-selection (or sorting) effects

• Growing body of literature on this topic

Latent Variables

• Green lifestyle propensity

• Luxury lifestyle propensity

Commute

Mode choice

(nominal)

Housing Type

(nominal)

Density of

Neighborhood (nominal)

Housing Cost

(grouped)

Average Commute

Distance (grouped)

Household

Vehicle

Type/Size

Number of

Bathrooms

(count)

Number of

Bedrooms

(count)

Unit-Square

Footage

(grouped)

Lot Size

(grouped)

Green Lifestyle

propensity

Luxury Lifestyle

propensity

Framework for Housing Choices and Activity Travel Behavior

Impact of Connected/Autonomous Transportation

• Safety enhancement• Virtual elimination of driver error – factor in 80-90% of crashes• No drowsy, impaired, stressed, or aggressive drivers• Reduced incidents and network disruptions• Offsetting behavior on part of driver

• Capacity enhancement• Platooning reduces headways and improves flow at transitions• Vehicle positioning (lateral control) allows reduced lane widths and utilization of

shoulders; accurate mapping critical• Optimized route choice

• Energy and environmental benefits• Increased fuel efficiency and reduced pollutant emissions• Clean fuel vehicles/Car-sharing

Impacts on Land-Use Patterns

Live and work farther away Use travel time productively Access more desirable and higher paying job Attend better school/college

Visit destinations farther away Access more desirable destinations for various activities Reduced impact of distances and time on activity participation

Influence on developers Sprawled cities? Impacts on community/regional planning and urban design

Impacts on Household Vehicle Fleet

Potential to redefine vehicle ownership No longer own personal vehicles; move toward car sharing enterprise where rental vehicles

come to traveler

More efficient vehicle ownership and sharing scheme may reduce the need for additionalinfrastructure Reduced demand for parking

Desire to work and be productive in vehicle More use of personal vehicle for long distance travel Purchase large multi-purpose vehicle with amenities to work and play in vehicle

Impacts on Mode Choice

Automated vehicles combine the advantages of public transportation with that oftraditional private vehicles Catching up on news Texting friends Reading novels

Flexibility Comfort Convenience

What will happen to public transportation?

Also Automated vehicles may result in lesser walking and bicycling shares

Time less of a considerationSo, will Cost be the main policy tool to influence behavior?

Impacts on Mode Choice

Traditional transit captive market segments now able to use auto (e.g., elderly, disabled)

Reduced reliance/usage of public transit?

However, autonomous vehicles may present an opportunity for public transit and carsharing Lower cost of operation (driverless) and can cut out low volume routes More personalized and reliable service - smaller vehicles providing demand-

responsive transit service No parking needed – kiss-and-ride; no vehicles “sitting” around 20-80% of urban land area can be reclaimed Chaining may not discourage transit use

COLLABORATE. INNOVATE. EDUCATE.

Individual attitudes regarding to autonomous

vehicles

There are several individual lifestyle, personality, and attitudinal factors that may impact the decision of owning/renting a connected/autonomous vehicle and use: Green lifestyle

Multitasking inclination

Tech-savvy people or geeks

Stressed drivers

For example, individuals who have a green lifestyle may search for locations that offer high accessibility to green areas,

may own fewer autos,

and may rent/ride autonomous vehicles (as public transportation or shared service) often.

The Bottom Line

Data to information – an important data science

Uncertainty, Uncertainty, Uncertainty

More uncertainty implies more need for analysis/planning

But analysis/planning must recognize the uncertainty (need achange in current thinking and philosophy)