Everybody loves New York City. Nobody likes car accidents. Why bother look at the motor vehicle collision data? Well, reality is reality. Road safety is by any means a critical issue, and is relevant to everybody’s daily life. It’s inevitable, and more often than not, a life-or-death situation indeed. Therefore, it is very important to look at the past collision history data and see what we can learn from the data to help better prevent and/or avoid collisions in the future. Meanwhile, this is a fairly challenging/interesting data-science problem by itself, which hence becomes the core motivation of this project.
The data set used is from the city government’s OpenData website, where a lot of useful data sets archived by city government are provided, including 311, 911 call history, restaurant inspection, traffic volume, traffic violation, etc. The NYC motor vehicle collision data set, contains up-to-date collision record ever since July 1st, 2012, and each record shows the date, time, location, the number of injured and/or killed people (both totally and in terms of pedestrian, cyclist, and motorist, respectively), along with the causes and involved types of vehicles, etc.
Looking at such a comprehensive data set, some interesting questions directly jumped into mind are:
- How is the data look like on a map? Can we find particularly more dangerous/risky regions of concern? What their collision history data look like, can we get any useful insight?…
- Especially for cases with walker injured/killed and the cases with people killed in general, are their distribution over location and/or time shows any significant different pattern/feature than the overall cases? Can we identify particular dangerous spots/areas for pedestrians, cyclists, or the lethal collisions, etc. ? …
- What are the top, say 20, most often seen collision causes and involved types of vehicles? What can learn from it? …
Also, some questions of general interest are:
- What is the trend of total number of collisions from year to year? What can we predict for 2018?
- What is the composition ratio of different types of victims (pedestrian, cyclist, motorist) and different levels of severities (no hurt, injured, lethal)?
- Was the situation different for different boroughs (5 boroughs of NYC)? What about different days in a week, hours in a day, month in a year? …
To address these questions, the specific project objectives are:
- Develop an interactive map tool to easily check and explore collision case info on a real city map,
- Conduct some preliminary exploratory data analysis to get overview results on interested questions.
Interactive Map Tool
To help easily visualize and explore the spatial details of the collision data, a comprehensive and flexible interactive map tool is developed using Leaflet package of R.
One particular nice feature of the tool is that: it can show heat map (collision data point intensity), cluster map (clustered collision data), and lethal collision markers (with detailed collision information pop-up) all at the same time. Besides, the user has highly flexible control on what portion of data he/she’d like to see, what year, borough, month range, and for what types of victims (pedestrians, cyclists, motorists) and what severity (no hurt, injured, lethal).
Preliminary Exploratory Data Analysis
Some preliminary analysis is done to get some high-level overview picture of the data set.
The figures below show the total number of collisions with respect to different years (and boroughs), month in a year, day in a week, and hour in a day, respectively.
Some major observations are as follows.
- Yearly-wise, there is an gradual collision increase since 2013, while a significant drop at 2016, but then increase again at 2017.
- Month, weekday, hour factors show results generally well aligned w/ common sense.
- Interestingly, Friday looks a little more like a peak day of a week in terms of number of collisions.
Particularly for the 2016 significant collision number drop, it’s mainly because of the successful Vision Zero campaign launched by the city government. How/what to predict for the situation of 2018 is definitely a fairly challenging/interesting problem deserving further/deep study/investigations…
Severity and Victims
The ratios of different severity levels and types of victims are shown below.
Some primary observations are:
- In most cases, collision are “No-hurt”, which is of much higher ratio than that of the “Injured” cases, while “Lethal” collisions are very rarely seen.
- In terms of victims, pedestrians consist of a significant portion of the total collisions, whose ratio is significantly higher than that of cyclists.
Especially for Manhattan:
- Interestingly, it has a much higher ratio of pedestrian victims than those of the other 4 NYC boroughs, mainly due to its special situation of highly crowded skyscraper buildings.
- Meanwhile, its “no-hurt” ratio is also much higher than those of the other boroughs.
It seems that the high pedestrian victim ratio may be the main contributing factor for the much higher no-hurt ratio. To confirm that, we need further investigate the no-hurt ratio of the pedestrian victims in Manhattan…
Causes and Involved Vehicles
The archived collision causes and involved types of vehicles are highlighted in the following frequency bar graphs. Note that herein to be more informative, we excluded the two most common top causes of “Driver Inattention/Distraction” and “Failure to Yield Right-of-Way“, and the two most common top involved types of vehicles of “Passenger Vehicle” and “Sport Utility / Station Wagon“, from the corresponding causes and vehicle graphs, respectively.
Some major observations are:
- Besides the commonly known reasons of bad driving habits/skills, a big portion of top causes are related to mental unconsciousness/fatigue/drowsiness, etc.
- Look at top involved vehicles, can find most of them are commercial vehicles.
With the above two observations, it looks like there may be existing reasonably high correlation between the two leading factors of drowsiness and commercial vehicle drivers. This would be a good direction for further study…
To facilitate the investigation/exploration of the collision data set, an interactive map tool is developed rendering comprehensive mapped information, along with flexible data control. Some preliminary analysis is done with some reasonable observations and interesting findings/thoughts for further study/investigation. Shiny app is available online. Source code is available at Github.
In summary, based on the top 20 causes and involved types of vehicles results, some good road safety advice are:
- Cautiously assume/take the right-of-way,
- Cautiously watch out for drowsy/commercial drivers.
As for city government, some reasonable ideas/suggestions coming out of the results are: to think of and define new/effective ways, and/or maybe stronger regulations, to better prevent drowsy driving (which is in effect may not be too much different than drunk driving actually), unsafe backing (some smart/effective way to better avoid this?), improper turning (and this?), etc. Hopefully, the more thinking/effort on these problems would help better prevent these high percentage causes of accidents in the future.
Based on findings so far, some interesting/promising directions to further pursue the topic include:
- Time series analysis to predict future trend
- Correlation analysis between drowsy and commercial drivers
- Correlation analysis w/ other data sets, esp bad/extreme weather data, and special event/celebration data, etc.
Finally, note that investigating this important data set has been a very hot topic during the past several years. There are quite many good/solid works/analysis/results out there for reference. This study is merely another personal journey at its very beginning…