5. Exploratory Data Analysis: A Note About Outliers - eliasmelul/CrimeInvestigation GitHub Wiki

Exploratory Data Analysis III

OtherTowed Figure 8. Daily Crimes for the 'Other' and 'Towed' offense groups over time

As we mentioned in Exploratory Data Analysis I: Top 15 Offense Groups there were notable outliers in some offense groups, most conspicuously the 'Other' and the 'Towed' offense groups. Figure 8 above shows the daily crimes time series for these offense groups and displayed some notable dates:

For the ‘Other’ offense group:

  1. Freq: 77; March 20th 2016
    • Red-Sox versus Mets
    • St. Patrick’s Day Parade
  2. Freq: 56, Jan. 4th 2018
    • Blizzard
  3. Freq: 32, Jan. 8th 2018
    • Blizzard

For the ‘Towed’ offense group:

  1. Freq: 88; March 13th 2018
    • Blizzard + Outages
  2. Freq: 64, Feb. 9th 2017
    • Blizzard + Outages
  3. Freq: 49, March 14th 2017
    • Blizzard

As we can observe, blizzards present opportunities for different crimes to occur sporadically. The Boston Police Department (BPD) should focus efforts investigating what effects of blizzards cause different types of robberies.

Since Blizzards are the most common cause for radical spikes in crime rates, possibly because of side-effects of blizzards, such as outages, we can predict that the 'Event' attribute in the dataset (denoting four levels: None, Snow, Rain, or Both) and possibly the 'Snowfall' variable are good predictors for the frequency of crime.