Download pptx - Car accident repairshops

Transcript

Data Collection – Vehicle Crashes Collected 1,048,575 records of vehicle crashes in New York from 2009 to 2012.(https://data.ny.gov/Transportation/Motor-Vehicle-Crashes-Case-Information-Beginning-2/e8ky-4vqe)

Queens

New York

Bronx

Hempstead

# of crashes in 2012(Group by municipality)

Day of Week vs. Time in a Day (Weekday) Vehicle crashes in weekday has two peaks in a day (about 8:00 and 17:00)

Day of Week vs. Time in a Day (Weekend) Vehicle crashes in weekend has only one peak in a day (after 12:00 pm)

Collision Type vs. Weather Condition The part of a car is hit the most under the normal weather condition (clear, cloudy, rain) are in Rear and Right Angle

Collision Type vs. Weather Condition The part of a car is hit the most under the unclear weather (snow, sleet, fog) have the same ratio of collision types

Data Mining

Problem: What factors cause multiple vehicle crashes

Output variables:

If this crash has more than 3 cars involved

Input variables:

Lighting Conditions

Road Descriptor

Traffic Control Device

Road Surface Conditions

Year, Day of Week, Time

Modeling: Bayes Point, Logistics Regression, Decision

Forest, Neural Network, SVM

Model Result – Bayes Point as an Example

RecallROC Curve

Prec

isio

n

Top 3 variables: Traffic Control Device, Time and Lighting condition

Variable Contribution

Model Result Comparison

SVM

Neural Network

Logits Regression

Decision Forest

Choose Logistic Regression as the optimal model

Lighting Conditions and Traffic Control Device are the most important factors

Variable Contribution

Number of Repair Shops in New York Collected 3,836 records of repair shops in New York