Today in NYC

• Approximately 4,000 people are seriously injured and over 250 are killed each year in motor vehicle accidents.
• Being struck by a vehicle is the leading cause of injury-related death for children under 14, and the second leading cause for seniors.
• On average, vehicles seriously injure or kill a New Yorker every two hours.
• 56% of all NYC traffic fatalities are pedestrians, versus 11% nationwide.

• An action plan released in 2014, in an effort to reduce traffic fatalities and injuries in NYC.
• Regard every accident as preventable incident that can be systematically addressed.
• A multifaceted approach, incorporating education, engineering, enforcement and legislation.
• 14 different Vision Zero initiatives that are in place today: Slowzones, Speed Humps, Signal Timings, Bike Priority, Enhanced Crossing, Lead Pedestrian Interval, Left Turn Traffic Calming, Safe Streets for Seniors, Priority Corridors, Zones and Intersections

• Exploratory study / Investigation of factors associated with the traffic incidents in NYC:

  • Who are involved in majority of traffic violations?
  • What are the Key contributing factors to the traffic accidents?
  • Where are the HotSpots?
  • Which boroughs had more accidents related to injuries/fatalities ?

• Impact of Vision Zero

  • Before and After Vision Zero analysis

• Time series analysis with traffic accident data

Note: Data for 2012, 2018 is incomplete

Note: Above are Kernal Density Graphs

• In the past, not having enough spatial data was a hurdle
• Today, too much data is the probelm !
• Too many scattered points on a map can be overwhelming !
• Displaying on a device would be processor intensive !
• Solution ?
  • Compress spatially redundant data points into a set of representative features.

• We need to provide K, the number of clusters.
• Place centroids \( C_1 ...C_K \) at random locations.
• Repeat until convergence: (Stop when none of the cluster assignments change)
  • for each point \( X_i \) :
    • find nearest centroid \( C_j \)
    • assign the point \( X_i \) to cluster j
  • for each cluster j = 1…K:
    • new centroid \( c_j \) = mean of all points \( X_i \) assinged to cluster j in prior step

• Human intuitive clustering method best suited for spatial data points (non-flat geometry/uneven clusters).
• Views clusters as areas of high density separated by areas of low density
• Got 2 parameters:
  • epsilon , essentially decides the size of the neighborhood.
  • minPoints, minimum cluster size
• You do not need to specify the number of clusters, it determines automatically based on the params.

Click here - HotSpot Viewer App

• Split the dataset - Persons injured in NYC Road Accidents before and after 2017
• Subset our dataset just for Brooklyn and Accidents caused by 'Driver Distraction'

• Summarized accidents by month.
• Train Dataset 55, Test dataset 15
• Explore this time series as a data visualization

• ARIMA is a model that can be fitted to time series data to better understand and predict future points in the series.
• ARIMA(p,d,q) model, where:
  • p is the number of autoregressive terms,
  • d is the number of non-seasonal differences needed for stationarity, and
  • q is the number of lagged forecast errors in the prediction equation. Let y denote the dth difference of Y, which means:
• If d=0: \( y_t = Y_t \)
• If d=1: \( y_t = Y_t - Y_{t-1} \)
• If d=2: \( y_t = (Y_t - Y_{t-1}) - (Y_{t-1} - Y_{t-2}) = Y_t - 2Y_{t-1} + Y_{t-2} \) (discrete analog of a second derivative)

• We used Dickey-Fuller test see whether the TS is Stationary or not: The test results comprise of a Test Statistic and some Critical Values for different confidence levels. If the 'Test Statistic' is less than the 'Critical Value', then the series is stationary.

• ACF (autocorrelation function) and PACF (partial autocorrelation) plots are used to find the AR(p) and MA(q) terms.

• Used log transformation of the series.
• To eliminate seasonality used Differencing which takes the difference with a time lag.

• Our model shows that there is a not much variation in the number of accidents due to DISTRACTED DRIVING from 2017 onwards.

• Mixed results from vision zero analysis.
• Human factors ( driver distraction, improper driving ) are major reasons for the accidents.
• Though Manhattan had more traffic violations, Brooklyn and Queens had more accidents that resulted in injuries/deaths.
• Over 50% of violations are attributed to younger drivers (35 and under).
• Passenger vehicles were most likely to be involved in the accidents.
• Hotspot viewer showed reduced pedestrian spots after VZ initiative. However no such indication in multi vehicle incidents.
• Time series analysis forecasted a gradual decrease in incidents.

• Include more incident data sets for hotspot analysis.
• Expand Time series analysis with other factors.
• Combine incident data with weather data and investigate the factors (environment factors like weather) for the increase of incidents in 2017.
• Enhance the incident data sets by imputing the location data.