Hire me

Work/PRJ-10/Exploratory analysis · World

Global terrorism, 1970–2017: where attacks happen, and how

Mostly in Iraq, Pakistan and Afghanistan, which had 46% of all attacks in 2012–2017. And mostly bombings: 49% of the 181,691 attacks from 1970 to 2017.

Data
Global Terrorism Database (START, University of Maryland): 181,691 attacks
Period
1970–2017 (1993 is missing)
Tools
Python, pandas, matplotlib
Code
github.com/BoredMongoose/Terrorism_analysis ↗
Business question

Where do terrorist attacks happen most, and what kind of attacks are they?

Recommendation

Put prevention effort where the attacks are: those three countries, and bombings first. In Central America and Sub-Saharan Africa, plan for armed assaults too, which are as common as bombings there.

Estimated impact46%

Of attacks in 2012–2017 happened in just three countries: Iraq, Pakistan and Afghanistan.

01Results

Where the attacks happened.

Where the attacks happened. Each cell counts the attacks at that spot from 1970 to 2017. The darkest area is Iraq; Pakistan, Afghanistan, northern India, the Andes and Central America follow.
Fig. 1Where the attacks happened. Each cell counts the attacks at that spot from 1970 to 2017. The darkest area is Iraq; Pakistan, Afghanistan, northern India, the Andes and Central America follow.
Attacks rose, fell, then shot up after 2011. The 2014 peak (16,903 attacks) came mostly from Iraq, Pakistan and Afghanistan. The bar for 1993 is missing because those records were lost.
Fig. 2Attacks rose, fell, then shot up after 2011. The 2014 peak (16,903 attacks) came mostly from Iraq, Pakistan and Afghanistan. The bar for 1993 is missing because those records were lost.
Bombings are almost half of all attacks. Bombing or explosion 49%, armed assault 23% and assassination 11%: together about 83% of attacks.
Fig. 3Bombings are almost half of all attacks. Bombing or explosion 49%, armed assault 23% and assassination 11%: together about 83% of attacks.

02How I did it

From raw data to a decision.

  1. Loaded the 181,691-row file with pandas and kept the dozen columns I needed.
  2. Counted attacks by year, country, region and attack type with groupby and value_counts.
  3. Checked the data: 1993 is missing, 45.6% of attacks have no known group, some casualty numbers are blank, and one longitude was impossible (−86,185,896), which I fixed by putting the decimal point back.
  4. Drew every attack's location as a density map, so the busiest places stand out.

03What could make this wrong

The limits, said out loud.

  • 01Different teams collected the data over the years, so some changes may come from how attacks were counted.
  • 02Blank killed and wounded numbers are counted as 0, so casualties are a little too low.
  • 03Two 2001 US rows look like one event recorded twice.

04What I learned

  • 01Check the data before trusting a chart: one longitude of −86,185,896 squashed my first map into a line.
  • 02A total hides the story: the 2014 peak came mostly from three countries, and the earlier rise from three different ones (Peru, El Salvador and Colombia).
  • 03Say what's missing: 45.6% of attacks have no known group, and 1993 isn't in the data at all.

05What I'd do next

  • 01Check the two 2001 US rows that look like one event recorded twice.
  • 02Look at attacks per million people, not just counts, so big countries don't dominate.
  • 03Add 2018 onwards when a newer version of the database is available.
See the code on GitHub ↗All projects