tahmid presentaion data_scienc_v2

8
Visualization and Analysis of 2015 Traffic Fatalities Tahmid Abtahi ,MS Student Computer Engineering, UMBC [email protected]

Upload: tahmid-abtahi

Post on 08-Feb-2017

18 views

Category:

Data & Analytics


0 download

TRANSCRIPT

Page 1: Tahmid presentaion data_scienc_v2

Visualization and Analysis of2015 Traffic Fatalities

Tahmid Abtahi ,MS StudentComputer Engineering, UMBC

[email protected]

Page 2: Tahmid presentaion data_scienc_v2

Dataset -2015 Traffic Fatalities Data • Annually Released by National Highway Traffic Safety Administration (NHTSA)• White House and U.S. Department of Transportation’s Blog calling for data

scientists, students and researcher to provide analysis on this data• Significance– Dept of Transportation aggressively seeking insights to improve road safety– Shaping Auto industry to improve vehicle safety– Identifying communities at higher risk of fatal crashes etc.– Insight to seek solutions to behavioral challenges like drunk, drugged, distracted and

drowsy driving

Page 3: Tahmid presentaion data_scienc_v2

Components of Data Set15 Tables

accident.csv – Crash data (State, county, Day, Hour, drunk driver, fatalities etc.)cevent.csv - Qualifying eventsdamage.csv - damaged area of vehicledistract.csv - driver distractiondrimpair.csv – physical impairmentnmimpair.csv – physical impairment of people not in vehiclesnmprior.csv - actions of non occupant peopleparkwork.csv - parked and working vehicles involvedpbtype.csv – Crashes between motor and pedestrians, bicyclistperson.csv - Person data file (age-sex-injury severity-air bag etc)vehicle.csv - Vehicle data type (number of occupants, model, make, registration state etc.)vevent.csv - sequence of event vindecode.csv vision.csv - circumstances which obscured driver vision

Page 4: Tahmid presentaion data_scienc_v2

Approaches• Data Clean up• Visualization• Applying classifiers

Results

Page 5: Tahmid presentaion data_scienc_v2

Results

Page 6: Tahmid presentaion data_scienc_v2

Results Given Month, Day and Hour - predicting a drunk driving related accident ?

Page 7: Tahmid presentaion data_scienc_v2

Next Ideas• Fuse multiple tables• Gender Bias on accident over state region• Distraction effects on fatalities• Clustering of pedestrian fatalities to identify potential risk regions

Source codes & visualizations in Kaggle kernels. Currently 3rd in the Top Contributor

Page 8: Tahmid presentaion data_scienc_v2

Questions?