Are we there yet? An Analysis of Cobb County Fire Department's - - PowerPoint PPT Presentation

▶
are we there yet
SMART_READER_LITE
LIVE PREVIEW

Are we there yet? An Analysis of Cobb County Fire Department's - - PowerPoint PPT Presentation

Sep 08, 2023 179 likes •555 views

Are we there yet? An Analysis of Cobb County Fire Department's Response Times. Ph.D. Students in Analytics and Data Science : Bogdan Gadidov , Lili Zhang, and Yiyun Zhou Faculty Advisors: Dr. Joe DeMaio, Dr. Kurt Schulzke, And Dr. Gene Ray

slide-1
SLIDE 1

Are we there yet?

An Analysis of Cobb County Fire Department's Response Times.

Ph.D. Students in Analytics and Data Science : Bogdan Gadidov, Lili Zhang, and Yiyun Zhou Faculty Advisors: Dr. Joe DeMaio, Dr. Kurt Schulzke, And Dr. Gene Ray

slide-2
SLIDE 2

Reducing Traveling Time for CCFD

2

  • Measured at the 90th percentile of the response times
  • National Standards: 4 mins
  • Cobb County: 8 mins
  • 29 fire stations and 272 fire zones
  • 168 incidents per day
  • Population 717,190 (50% increase since ‘95)

22 Getting emergency vehicles quickly to an incident is critical in saving lives and property.

2

slide-3
SLIDE 3

You may have noticed…

3 3

slide-4
SLIDE 4

4 4

slide-5
SLIDE 5

5 5

slide-6
SLIDE 6

Reducing Roll time for CCFD

6

Historically, fire zones were created over the past few decades by driving around in a vehicle and eyeballing where zones should begin and end.

6

slide-7
SLIDE 7

Historical Fire Station Zones

7 7

slide-8
SLIDE 8

Reducing Roll time for CCFD

8

Population growth and surrounding infrastructure has changed dramatically in Cobb County over the past two decades. Town Center Mall in the late 80s.

8

slide-9
SLIDE 9

Reducing Roll time for CCFD

9

Population growth and surrounding infrastructure has changed dramatically in Cobb County in the past decades. Town Center Mall more recently.

9

slide-10
SLIDE 10

10 10

Objective

  • Find the most important variables that can

influence the traveling time of CCFD

  • Goal of CCFD: traveling time within 4 minutes

90% of the time.

  • Then, see if the variables can be combined with

Google maps to optimize the zones.

slide-11
SLIDE 11

11 11

Variables from the datasets provided by clients

  • Population:

< 1st Quantile (<2299): pop = 1; 1st ~ 2nd Quantile (2299~4278): pop = 2; 2nd ~ 3rd Quantile (4278~6481): pop = 3;

  • therwise: pop = 4
  • Fire (Yes/No)
  • Medical (Yes/No)
  • Rescue (Yes/No)
  • Unit (Vehicle Type)
slide-12
SLIDE 12

12 12

  • Unit (Vehicle Type)
slide-13
SLIDE 13

13 13

Variables from the dataset collected online

  • Temperature:

< 32 0F: Cold; °F 32 ~ 59 0F: Freeze; 59 ~ 85 0F: Warm; > 85 0F: w_Hot.

  • Visibility (in miles):

> 3: indvisib = 1; <= 3: indvisib = 0.

slide-14
SLIDE 14

14 14

Variables from the dataset collected online

  • Day_of_Week: 1, 2, 3, 4, 5, 6, 7
  • Interval_of_day:

5 ~ 10 AM: Morning 10 ~ 16 PM: Mid 16 ~ 20 PM: Evening Otherwise: Night

slide-15
SLIDE 15

15 15

Variables from the dataset collected online

  • Events
slide-16
SLIDE 16

16 16

Variables created by meaningful interactions

  • Rain and Interval of day
  • Snow and Interval of day
  • Population and Interval of day
slide-17
SLIDE 17

17 17

Response Variable (traveling time)

  • Delete observations with traveling time > 95th percentile

since these observations are potential outliers;

  • Traveling time <= 4 min then GOODBAD = 0, denoting the

success to meet the goal;

  • Traveling time > 4 min then GOODBAD = 1, denoting the

failure to meet the goal.

slide-18
SLIDE 18

18 18

Model Performance: Area under ROC

Receive Operating Characteristic Curve (ROC curve): a graphical plot that illustrates the performance of the classification method. X-axis: false positive rate Y-axis: true positive rate Decision Tree: 0.636 Logistic Regression: 0.674 Random Forest: 0.639

slide-19
SLIDE 19

19 19

Model Performance: Misclassification Rate

slide-20
SLIDE 20

20 20

Directions of Variable Effects

Variable Name Influence on Traveling Time

Pop_eve Increase Pop Increase Pop_mid Decrease Vehicletype (= MEDOP) Decrease Fire (= Yes) Decrease Humidity Increase Pop_morn Decrease Temperature Decrease Medical (= Yes) Decrease Visibility Decrease Other (= Yes) Increase Events (= Snow-Thunderstorm) Increase

slide-21
SLIDE 21

Reducing Traveling Time for CCFD

21

Cobb County – Historical Traveling Time Data Google Maps – New Traveling Time Data

21

slide-22
SLIDE 22

Reducing Traveling Time for CCFD

22 22

slide-23
SLIDE 23

Algorithm Implementation:

23

Step 1. Retrieve relevant information of all incidents.t

  • Address
  • Zone
  • Assigned station
  • Traveling time of the unit that responded
  • Time interval

23

slide-24
SLIDE 24

Time Interval For Each Day

24

5 : 00 to 10 : 00 Morning 10 : 00 to 16 : 00 Mid 16 : 00 to 20 : 00 Evening 20 : 00 to 5 : 00 Night

24

slide-25
SLIDE 25

Algorithm Implementation

25

Step 2. Retrieve the addresses of all 29 Cobb County fire

  • stations. best station suggested by Google Maps to the historical corresponding station.

STATION ADDRESS 1 5656 Mableton Pkwy SW, Mableton, GA 30126, USA 2 208 Barber Rd, Marietta, GA 30060, USA 3 580 Terrell Mill Rd, Marietta, GA 30067, USA 4 1901 Cumberland Pkwy SE, Atlanta, GA 30339, USA 5 4336 Paces Ferry Rd SE, Atlanta, GA 30339, USA 6 5075 Hiram Lithia Springs Rd SW, Powder Springs, GA 30127, USA 7 810 Hurt Rd, Austell, GA 30106, USA 8 2380 Cobb Pkwy NW, Kennesaw, GA 30152, USA 9 7300 Factory Shoals Rd, Austell, GA 30168, USA

25

slide-26
SLIDE 26

Algorithm Implementation

26

Step 2. Retrieve the addresses of all 29 Cobb County fire

  • stations. best station suggested by Google Maps to the historical corresponding station.

Ideally, we would check all 29 fire stations for their Google Maps travel time to each incident. Unfortunately Google Maps access is only free up to 2,500 requests per day. station

suggested by Google Maps to the historical corresponding station.

Premium access is available at 50 cents per 1000 requests with a limit of 100,000 requests per day..

26

slide-27
SLIDE 27

Algorithm Implementation

27

Step 3. Create a list of neighoring stations to check for each incident.

27

slide-28
SLIDE 28

Algorithm Implementation

28

Step 4. Use Google Maps Distance Matrix API to calculate traveling time from an incident to its neighboring stations at midday for 2016.

Zone Hist_time Corr_statio n Google_stati

  • n

Google_ti me 24E 515 24 24 281 26C 192 26 26 196 22D 287 22 22 385 1C 195 1 1 284 9C 243 7 1 47 22F 160 22 22 143 15B 402 16 14 362 5B 304 5 5 385 27C 156 27 27 119

28

slide-29
SLIDE 29

Python Code To Retrieve Google Data

29 29

slide-30
SLIDE 30

Algorithm Implementation:

30

Step 5. Reshape Zones. Note that zone 8 is Kennesaw State University which exploded in population and infrastructure over the past two decades.

30

slide-31
SLIDE 31

31

Travel Time Reduction

Zone Old Time

New Station New Time Time Savings

17K 521 10 383 138 8I 490 24 375 115 26G 631 11 517 114 8H 521 17 419 102 19F 318 3 222 96 18F 497 3 412 85 11B 345 28 319 26 18A 321 24 298 23 1C 269 27 260 9

31

slide-32
SLIDE 32

32

Limitations

  • Google Time impacted by current traffic outliers
  • I-85 Collapse
  • Unusually large accidents
  • Bad weather
  • ??????
  • Units may not drive from station to incident location
  • Available data from Google Maps
  • Slightly under $1000 to run all incidents from 2011

during midday

32

slide-33
SLIDE 33

33

Future Work

  • Reshape zones for other times of day
  • Morning
  • Evening
  • Night
  • Reshape zones for summer when school is out
  • Reshape zones for Braves games
  • Use Premium API key or low-tech parallel processing

33

slide-34
SLIDE 34

34 34

slide-35
SLIDE 35

35