---

[Project Summary] [Project Goals] [Deliverables] [Project Planning] [Initial Hypothesis] [Equipment Rail Data Dictionary] [Highway Rail Data Dictionary] [FIPS State Codes] [DS Pipeline] [Conclusion] [Next Steps] [Reproduce This Project] ​ ​

Project Summary:

Our team accessed the U.S. Department of Transportation’s database to analyze 8 years of rail accidents across the United States. We utilized the full data science pipeline to analyze the data and create a classification model that would predict which company would be involved in a rail accident.

Data Source: U.S. Department of Transportation

---

Project Goals:

• Determine which features are drivers of highway rail accidents and equipment rail accidents.
• Build a classification model for predicting which railroad operator is most likely to be involved with a given accident. This information is used to enhance the overall analysis.

[Back to top]

---

Deliverables:

• Showcase highlighted findings in a presentation delivered to stakeholders.
• Create a reproducible jupyter notebook report that includes process, takeaways, and discoveries from every stage of the pipeline.

[Back to top]

---

Project Planning:

All of our project planning can be found on Trello. Below is a link to our progress along with a visual snapshot.

A link to the Trello board below can be found at https://trello.com/b/msJXyeEv/off-the-rails

Here is a snapshot of our project planning/setup on the evening of 5/23/21

[Back to top]

---

Initial Hypotheses:

• For each type of rail incidents, Railroad Operator plays into the frequency of incidents and the severity of incidents.
• Geography and location has an impact on the type of rail incident.
• Time of year and weather conditions have an impact on the frequency and scale of a rail accident.
• There will be a difference in severity of accident based on whether the accident was a highway rail accident or an equipment accident.

[Back to top]

---

Data Dictionary

* - Indicates the target feature in this data

Equipment Data

Features	Description	Data Type
railroad_company *	railroad code	object
accident_type	type of accident	int
state	FIPS state code	int
temp	temp in degrees	int
visibility	encoded visibility	int
weather	weather conditions	int
train_speed	speed of train in mph	int
train_direction	train direction	float
train_weight	gross tonnage of train	int
train_type	type of equipment	object
track_type	type of track	int
front_engines	# of head end locomotives	int
loadfrght_cars	# of loaded freight cars	int
loadpass_cars	# of loaded passenger cars	int
emptyfrght_cars	# of empty freight cars	int
emptypass_cars	# of empty passenger cars	int
equip_damage	equipment damage is USD	int
track_damage	track damage in USD	int
cause	cause of incidence	object
total_killed	# of killed	int
total_injured	# of injured	int
max_speed	maximum equipment speed	int
total_damage	total damage	int
engineers_onduty	# of engineers on duty	float
conductors_onduty	# of conductors on duty	float
brakemen_onduty	# of brakemen on duty	float
region	FRA designated region	int
typrr	type of railroad	object
lat	latitude	object
long	longitude	object
signal_type	type of signal	int
date	date of incident	datetime
season	season of year	object

[Back to top]

---

Highway Data

Features	Description	Data Type
railroad_company *	railroad code	object
station	nearest timeable station	object
county	FIPS county code	object
state	FIPS state code	int
region	FRA designated region	int
city	FIPS city code	object
vehicle_speed	estimated vehicle speed	float
vehicle_type	encoded highway user	object
vehicle_direction	encoded user direction	object
position	encoded user direction	object
accident_type	circumstance of accident	int
hazmat_entity	entity transporting hazmat	object
temp	temperature in degrees	int
visibility	encoded visibility	int
weather	weather conditions	object
train_type	type of equipment	object
track_type	type of track	object
front_engines	# of head end locomotives	int
railcar_quantity	quantity of railcar	int
train_speed	estimated train speed	float
train_direction	direction of train	object
warning_location	location of warning	object
warning_signal	crossing w highway signal	object
lights	crossing illuminated	object
standveh	passed a standing vehicle	object
other_train	another train involved	object
motorist_action	action of highway user	object
view_obstruction	track view obstruction	int
vehicle_damage	vehicle damage in USD	float
driver_fate	fate of driver	object
vehicle_occupied	was the vehicle occupied	object
total_killed	total # of deaths	int
total_injured	total # of deaths	int
vehicle_occupants	# of vehicle occupants	int
ispublic_crossing	is this a public crossing	object
fips	FIPS code	int
whistle_ban	whistle ban in effect	object
driver_age	age of driver	object
driver_gender	gender of driver	object
train_occupants	# people on train	int
user_killed	# of drivers killed	int
user_injured	# of drivers injured	int
rail_killed	# of rr employess killed	int
rail_injured	# of rr employees injured	int
train_pass_killed	# train passengers killed	int
train_pass_injured	# train passengers injured	int
road_conditions	encoded road conditions	object
date	date of incident	datetime
season	season of year	object

[Back to top]

---

FIPS State Codes

FIPS Code	Corresponding State
01	Alabama
02	Alaska
04	Arizona
05	Arkansas
06	California
08	Colorado
09	Connecticut
10	Delaware
11	District of Columbia
12	Florida
13	Georgia
15	Hawaii
16	Idaho
17	Illinois
18	Indiana
19	Iowa
20	Kansas
21	Kentucky
22	Louisiana
23	Maine
24	Maryland
25	Massachusetts
26	Michigan
27	Minnesota
28	Mississippi
29	Missouri
30	Montana
31	Nebraska
32	Nevada
33	New Hampshire
34	New Jersey
35	New Mexico
36	New York
37	North Carolina
38	North Dakota
39	Ohio
40	Oklahoma
41	Oregon
42	Pennsylvania
44	Rhode Island
45	South Carolina
46	South Dakota
47	Tennessee
48	Texas
49	Utah
50	Vermont
51	Virginia
53	Washington
54	West Virginia
55	Wisconsin
56	Wyoming

[Back to top]

---

Data Science Pipeline:

1. Acquire

• The data is acquired from csv files sourced from the US Department of Transportation.
• Two dataframes are created by concatenating the csv files for highway rail accidents and equipment rail accidents.
• All functions to prepare the data are included in acquire.py.

2. Prepare

• Prepare the data for analysis with prepare.py module with functions for each type of rail accident (highway and equipment)
• The prepare module will return respective dataframes, split into train, validate and test. The train dataset will be ready for exploratory analysis.

3. Explore & Preprocessing

• Run univariate, bivariate, and multivariate visulaizations for how features interact with each other and the target, railroad_company
• Run statistical tests to help answer questions that arose from exploration
• All functions to explore the data are included in explore.py.
• Preprocessing module prepares the dataframes for modeling. All functions for preprocessing are included in preprocessing.py.

4. Model/Evaluate

• Develop a baseline model for predicting the railroad operator based on the type of incident and incident features
• Build classification models that improve upon the baseline accuracy with the understanding that model performance and evaluation metrics will better inform the overall analysis.

5. Deliver

• A deployed model that can take in new data that is preprocessed and returns similar results. This is important for supporting any analysis and takeaways that were concluded from the modeling stage.
• A report with visuals that highlight the findings from the project analysis
• A reproducible notebook that is well-documented

[Back to top]

---

Conclusion:

• US Railroads seem to be operating in a consistent manner with regards to Equipment Rail Accidents.
• Characteristics of the highway driver and location, rather than the Railroad, appear to account for the variance in outcomes of highway rail accidents.

Next Steps:

• Industry and local governments should further investigate the drivers for highway rail accidents, especially in locations with recurring incidents.

[Back to top]

---

Instructions for Reproducing Project:

1. Read and follow this README.md.

2. Download the following files to your working directory:

• acquire.py
• prepare.py
• explore.py
• preprocessing.py

3. Run our final Jupyter Notebook (final_report_and_findings.ipynb) to reproduce our findings and analysis.

[Back to top]

---