Data

The global Terriorism dataset consists 181691 rows and 23 columns. Each row corresponds to the information of a terrorism incident that happened around the world. The dataset records terrorism incidents dated from 1970s to year 2017. The main variables we are using are as following:

  • Location (categorical) is accurate to name of the city, village, or town in which the incident occurred.

  • Time of event (qualitative) is accurate to the day in which the incident occurred.

  • Type of attack (categorical) is the general method of attack, such as assassination, bombing, hostage etc.

  • Type of weapon (categorical) is the main weapon used, such as Firearms, Explosives etc.

  • Motive (text) is a column explaining the cause of the event, such as intimidate the police, protest Vietnam War, fight imperialism etc.).

  • Ransom (text) is a column explaining any information about non-money demands made by perpetrators, as well as information on conflicting reports of how much money was demanded and/or paid.

  • Other interesting variables we will look into include, longitude & latitude (qualitative), attack target (categorical, such as Government, Police etc.).

Problem

The main research project problems we would like to learn are:

  • Which areas are more susceptible to terrorism in the world? Are the vulnerable places changed through years?

  • How do the weapon type and the target type vary in the terriorsm attacks from years 1970 to 2017? If so, are there any generalized trends? Are there any specific weapons that certain terrorism groups would use for specific target type?

  • What are the common motivation and demand among the terrorism attacks?

  • Case Study: What are the most common terrorism groups and attack locations in India, which is one of the most vulnerable contries? How do these attacks’ severity vary across India? What is the scale of the attack in terms of the number of the terrorists involved?

Methods

The following are the approaches we make to solve the problems.

Firstly, we would like to learn which countries or areas are more susceptible to terrorism by examining the number of terrorism attacks across the world with a choropleth map.

We can spot some main areas where terrorist incidents occur the most over time. The top places are Pakistan, Afriganistan, Iraq, and Nigeria. East Asia and America show much fewer incidents, compared to places above.

In order to learn the terrorism attack trend from the years 1970 to 2017 from different regions of the world, we focus on variables iyear, region_txt, n_events in a time series plot.

Overall, there is an increasing trend in the number of incidents by year in the world. After the year 2000, the number of terrorism events rapidly increases in the Middle East & North Africa, and South Asia, and sub-Saharan African. During 1980 and 1997, South America has most terrorism events. The number of terrorism events in all regions decreased to the lowest points around 1997. After the year 2015, terrorism events decreased rapidly in the Middle East & North Africa and South Asia.

In addition, we want to understand the variability of target type over time and what are the more heavily used types of weapons used on certain targets. Therefore, we need to look at the marginal distribution as well as the conditional distribution of variables Target and Weapon on the timeline.

The above graph suggests that overall the number of terrorism for each type of events is increasing. we also see an unusual spike in the number of terrorism events in both Military and Police after 2010. Business is also a major target type. The most defining feature of the “seasonal” plot is that after 2010, Military and Police has large variation than other target types. This is likely due to the spike Military and police after 2010. Between 1995 and 2005 year, Religious figures and Institutions have more variation. 1980 -1990 year, Utilities exhibits much more variation. The business also has some variations before the 1980 year. There is a gap in the 1994 year that all targets have few variations. Finally, there does not appear to be any clear trend in the “irregular” plot.

Since we can see the clear trend in Military, Utilities, Police and Business from graphs above. We want to know the autocorrelation among these four targets.

There are clearly seasonality trends for four groups. Military and Business seem to peak and valley at similar times. There is a much higher positive autocorrelation for lags 0 to 100 in those four groups. Perhaps the most noticeable thing is that autocorrelations “last longer” for Police than other groups. The autocorrelation tends to switch from positive to negative (and vice versa) at further time points for Police in lags 0 to 100 and lags 300 to 600. lags around 400 in the Utilities graph are not significantly different from zero.

After carefully examine the target type of attacks alone, we want to find some correlation between target type and weapon type by making a conditional bar plot.

By looking at the bar plot above, from the distribution of weapon type 1 conditioning on target type 1, we can see that explosives and firearms take large part across all tart types. Private Citizens and Property are the most targeted object among all.

Moreover, we would like to know the most common motive (e.g. religious sentiments, personal grudges) and demand among these terrorism attacks by plotting the word clouds on motive and ransom text.

We can see that the most frequent words in motive are “victim”, “military”, “part”. A lot of words related to religous wars pop up in the wordcloud, such as “sunni”, “sectarian”, “iraq”, “extremist”, “taliban”, “muslim”. Other frequent words often involve political combats, such as “maoist”, “communist”, “polic”.

Ransomnote text is about the requests made by the perpetrators. Large part of ransom is about “hostage”, “kidnap”, money “exchange”, and releasing “prisoner”. Also, we see that many terrorist incidents did not request ransom at all, with “unknown” denoted.

Finally, we conduct a case study on India, one of the most volunerable countries in the world. We want to know more about the terrorism attacks in India, especially the most common terrorism groups, the most prevalent activity locations and the scale of attacks with a treemap on India subset.

This treemap chart provides a hierarchical view of the terror groups and their activity locations. Meanwhile, the color represents the number of killings in the terrorism events, and the size represents the number of perpetrators attended.

It suggests that India faces a wide range of terror groups, including terrorism for political goals(e.g., CPI-Maoist, Maoists), religious goals(e.g., Hindu extremists) and ideological goals(e.g., GNLF). Among them, political terrorism groups dominate almost 2/3 of all terrorism events, such as the CPI-Maoist and the Maoists.The location of terrorism events concentrate in Bihar, Jharkhand and Chhattisgarh. Although these 3 regions have the most number of perpetrators involved terrorism events, Jammu and Kashmir have the most killings.

Conclusion

In conclusion, we found out throughout our analysis that there’re indeed certain parts of the world that are more vulnerable to terrorist attacks than others. In addition, we discovered some trends in the number of attacks over the years and with respect to seasonal change. We also found some weapon types that are more frequently used against certain targets. By analyzing text we found the top motivations and reasons for the attacks.

For the India case study, we discovered the major terrorist groups, the locations from which they usually operate, and the scale of attacks for these groups.

Future Work

In the future, we would like to do more case study in other countries in order to find more interesting patterns, taking into account the national conditions. In addition, we would update the data from year 2017 to 2020, having more insights about the recent trends. We might study the relationship between the number of terrorism attacks and the effect of population growth by fitting regression models.