Summary & Index:

In this weeks we conducted the final steps of our descriptive tasks: analysis of terrorism perpetrators. The main conclusions are:

• Around half of the attacks have no identified perpetrator.
• It is possible to cluster the terrorist groups based on target type, attack type, and weapons used.
• In the next weeks, we should do some tasks of predictive analysis.

Wiki Status - concluded

Index

###1 - Terrorism perpetrators description ####1.1 - Research Questions ####1.2 - Exploratory Clustering of Terrorist groups ####1.3 - Geo-Referential Exploratory Analysis ###2 - Conclusion and future steps ###3 - Weekly Presentation ###4 - Perceived Feedback

Weekly work

1 - Terrorism perpetrators description

1.1 - Research Questions

In this weeks we focus on different research questions in order to describe and better understand the terrorist perpetrators:

• Who are the terrorism perpretators? Where are they from? Do they act in groups? Are they working alone?
• Which kind of attack do they use? Does the type of attack depend on the group?
• Which kind of weapons do they use? Does the type of weapon depend on the group or on the type of attack?
• Do they commit suicide? Does commiting suicide depends on the type of attack or group?
• Are they captured? Does this depend on the region?
• Are there terrorist groups more successful in their attack?

We present here the answers found.

Do they act in organized groups?

Regarding to the perpetrator(s) of the terrorist attacks we found:

• 47.25% are "Unknown";
• 0.56% are classified as either "Other" or "Unaffiliated Individual(s)";
• 52.19% are terrorist groups.

Based on this, we can conclude that in ~48% of the attacks it's not possible to find out who is the responsible for it.

Besides, ~52% of the terrorist attacks are conducted by terrorist groups. In order to better understand and study these groups, we analyzed the union set of the top 30 groups with more incidents, more killings and active during the largest periods of time for each region. This originated a list of 513 different terrorist organizations. From this list we excluded the terrorist group names that are vague (e. g. Palestinians), ending with a list of 403 names.

Who are they?

In fig. 1 it's possible to see the top 15 groups with more incidents. The Taliban and Shining Path (Communist Party of Peru) are the most noticeable groups, both with more than 4000 incidents confirmed.

#####Figure 1 - Top 15 groups with more incidents (click on images to open in new tab)

Where are they from?

Regarding the countries with more incidents registered, fig. 2 indicates that Afghanistan, Peru and Colombia distantiate from the other ones, having more attacks. This could be related to the high number of incidents from the groups acting in these countries, the Taliban (Afghanistan) and Shining Path (Peru), which in fig. 1 are also the more active groups.

#####Figure 2 - Top countries with more incidents

Do they act in groups? Are they working alone?

Regarding the median number of perpetrators, Central America and Caribbean have the highest value (15). However, it is also interesting to notice that Australasia & Oceania has a high median value of perpetrators per attack, despite the low number of attacks (see fig. 3) in this area.

#####Figure 3 - Top Regions with median highest number of perpetrators in one attack

Which kind of attack do they use? Does the type of attack depend on the group?

In fig. 4 it's possible to notice that bombings and armed assaults are the most frequent types of attacks. When analyzing this frequency in the top 15 groups with more incidents, the pattern is still visible (fig. 5). However, there's variability between the groups, especially regarding the percentage of Armed Assaults, Assassinations, Bombings and Hostage Takings.

#####Figure 4 - Number of terrorist attacks for each type.

#####Figure 5 - Types of attack in each terrorist group.

Which kind of weapons do they use? Does the type of weapon depend on the group or on the type of attack?

In fig. 6, we can notice that explosives and firearms are the most used weapons. When analyzing this frequency in the top 15 groups with more incidents, the pattern is still visible. Most of the groups have at least 50% of their attacks being associated with explosives (especially the Islamic State) (fig. 7).

#####Figure 6 - Frequencies of types of weapons used.

#####Figure 7 - Types of weapons in each terrorist group.

When analyzing the association between the attack types and the weapon types (fig. 8) it's possible to identify 1 cluster of similar types of attacks, regarding the weapon type ratios. The types that seem to belong to this cluster are Hostage Taking, Hijacking, Assassination and Armed Assault.

#####Figure 8 - Interaction between the attack types and the weapon types.

Do they commit suicide? Does commit suicide depends on the type of attack or group?

By looking at the data, we concluded that in 4% of the attacks at least one perpetrator committed suicide. In fig. 9 it's visible that the type of attacks that involve more suicides are:

• Bombings
• Hostage Takings
• Hijacking (of vehicles)
• Assassinations

#####Figure 9 - Percentage of suicide by attack type.

Regarding the type of weapons involved in suicidal attacks (fig. 10), we concluded that Vehicles (excluding car bombs or truck bombs) and explosives have the greatest rates of suicides.

#####Figure 10 - Percentage of suicide in each weapon type.

In terms of terrorist groups, fig. 11 illustrates the top 15 groups with more than 1 incident that attack more with suicide. We conclude by manual inspection that almost all of them operate in the Middle East region.

#####Figure 11 - Top 15 groups that attack more with suicide.

Are they captured? Does this depend on the region?

In terms of capture rates (fig. 12), we can see that North America and East Asia have the higher values (> 0.7). On the other hand, Sub-Saharan Africa, Southeast Asia, South Asia, Eastern Europe and South America have the lowest values (< 0.5).

#####Figure 12 - Percentage of terrorist captures in each region.

Are there terrorist groups more successful in their attack?

Regarding the success rates, we concluded that 32% of the terrorist have a success rate higher than 95%. In fig. 13 when can observe the 10 terrorist groups with 5 incidents or more, which indicates that usually, the terrorist have a high percentage of successful attacks, with only 2 groups having a success rate less than 50%

#####Figure 13 - Groups with lower success rates.

In the future analysis, we can check if these terrorist groups verify other common properties and belong to one cluster.

1.2 - Exploratory Clustering of Terrorist groups

Using the reduced set of interesting terrorist groups (N=403), we performed clustering based on various parameters:

• Clustering methodology - K-means clustering.
• Processing of raw data - 1-hot-encoding of the categorical variables (weapontype[1-4], targettype[1-3], attacktype[1-3]) was done and a profile was created for each group based on the frequency of each type.

The clustering algorithm was applied to the group profiles with different values of k, to obtain meaningful cluster centers. Several different approaches were tested. We present the results found in the next subchapters.

1.2.1 Clustering based on Weapons Used

In the fig. 14 we can see that the groups can be classified into 3 major categories based on the type of weapons more used.

#####Figure 14 - Clusters found based on Weapons used.

The clusters formed are:

a. Firearms Users: Boko Haram, Communist Party of India - Maoist (CPI-Maoist), New People's Army (NPA), Kurdistan Workers' Party (PKK), Liberation Tigers of Tamil Eelam (LTTE).

The major areas where these groups (as seen in the video below) operate are South Asia, Central and Middle East Asia, Central Africa and Eastern Europe.

b. Explosives Users: Taliban, Shining Path (SL), Islamic State of Iraq and the Levant (ISIL), Farabundo Marti National Liberation Front (FMLN), Revolutionary Armed Forces of Colombia (FARC).

The major areas of operation for these groups are South America, Middle East and South Asia and Europe.

c. Incendiary Users: Sandinista National Liberation Front (FSLN), Muttahida Qami Movement (MQM), Animal Liberation Front (ALF), Earth Liberation Front (ELF), Jamaat-E-Islami (Bangladesh) (Number of events for this category is low ~100)

Majority of these attacks can be observed in North America, Europe and Bangladesh.

We try to test if this clusters are uniformly distributed all over the globe (Video 1).

#####Video 1 - Geographical distribution of the clusters based on Weapons used.

In general the three clusters are common all over the globe. However, we can see that Incendiary Users are less common in South America.

1.2.2 Clustering based on Attack type

In the fig. 15 we can see that the groups can be classified into 4 major categories based on the type of Attack more frequent.

#####Figure 15 - Clusters found based on Attack type.

The clusters formed are:

a. Bombers: Shining Path (SL), Islamic State of Iraq and the Levant (ISIL), Farabundo Marti National Liberation Front, Basque Fatherland and Freedom (ETA), National Liberation Army of Colombia (ELN) Frequent geographical regions where such attacks are conducted include Middle, South and South East Asia, South America and Europe.

b. Armed Assaulters: Taliban, Boko Haram, Communist Party of India - Maoist (CPI-Maoist), Kurdistan Workers' Party (PKK), Liberation Tigers of Tamil Eelam (LTTE).

Geographically these groups attack regions in Middle East and South Asia, East Europe, and South and Central Africa

c. Mixed (Most frequent style - Assassinations): Revolutionary Armed Forces of Colombia (FARC), Irish Republican Army (IRA), New People's Army (NPA), Al-Shabaab, Al-Qaida in the Arabian Peninsula (AQAP).

Most groups target Europe (especially Northern Europe), East Africa, East Asia and Northern South America

d. Infrastructure/Facility Attackers: Animal Liberation Front (ALF), Earth Liberation Front (ELF), United Front for Democracy Against Dictatorship, Independent Armed Revolutionary Commandos (CRIA), Senki ("Battle Flag") (Number of events is low ~100).

Majority of these groups operate in Europe and North America.

We try to test if this clusters are uniformly distributed all over the globe (Video 2).

#####Video 2 - Geographical distribution of the clusters based on Attack type.

In general, the four clusters are common all over the globe. However, this is not the case with the Infrastructure/Facility Attackers. This kind of attackers prefers the North America and Central Europe.

1.2.3 Clustering based on Target type

In the fig. 16 we can see that the groups can be classified into 6 major categories based on the target type that they prefer more.

#####Figure 16 - Clusters found based on target type.

The clusters formed are:

a. Police: Taliban, African National Congress (South Africa), Al-Gama'at al-Islamiyya (IG), Algerian Islamic Extremists, Islamic Salvation Front (FIS) (Private Citizens and property are the secondary point of attack for this group, followed by Government(General) and Businesses)

These groups attack regions in South and Central Asia, Southern Africa and pockets of Northern Africa. There are also some sporadic events in North and South America.

b. Business: Basque Fatherland and Freedom (ETA)', 1905), ('Corsican National Liberation Front (FLNC)', 627), ('Tupac Amaru Revolutionary Movement (MRTA)', 522), ('Dev Sol', 233), ('Red Brigades', 212)] (Businesses are the secondary point of attack for this group, followed by Government(General) and Private Citizens and property)

Majority of these groups can be seen to coordinate attacks in Europe and North America, followed by South America.

c. Mixed (Most frequently attacks Citizens): Shining Path (SL), Farabundo Marti National Liberation Front (FMLN), Revolutionary Armed Forces of Colombia (FARC), Irish Republican Army (IRA), Communist Party of India - Maoist (CPI-Maoist) (Government(General), Businesses and Police are also often attacked).

This cluster of groups have a wide geographical spread with many attacks all over Asia, Africa, Europe and South America.

d. Private Citizens and Property: the Islamic State of Iraq and the Levant (ISIL), Boko Haram, Al-Qaida in Iraq, Death Squad, Moro Islamic Liberation Front (MILF) (They also sometimes attack Governments and Businesses).

Central Africa and the Middle East see the highest frequency of attacks from groups in this cluster.

e. Government (General): Independent Armed Revolutionary Commandos (CRIA), Popular Militia (Colombia), Supporters of Ex-President Gamsakhurdial, Tamil Nadu Liberation Army, All Burma Students' Democratic Front (ABSDF) (Low number of events ~10)

Due to the low number of events, we only see a few attacks in Sri Lanka, Myanmar, Japan and Colombia and Puerto Rico.

f. Government (Diplomatic): Justice Commandos for the Armenian Genocide, Croatian Nationalists, Islamic Jihad Group (IJG), Maruseido (Marxist Youth League), Union of Congolese Patriots (UPC) (Also attacks Private Citizens and Property. Number of incidents are low ~10)

There are very few events in this cluster, the majority of which occur in Eastern Europe followed by a few in various parts of the world.

We try to test if this clusters are uniformly distributed all over the globe (Video 3).

#####Video 3 - Geographical distribution of the clusters based on Target Type.

In this case the six clusters are clearly happening in different regions of the globe. In further phases this will be analysed more in depth.

1.2.4 Conclusion

The exploratory clustering analysis allowed us to conclude that several variables can be used for the clustering of terrorist groups. The results seen so far prove that terrorist groups use different methods of attack and act in different places. In the future, we will try to combine these variables at the same time and check if more clustering models arise.

1.3 - Geo-Referential Exploratory Analysis

With the Geo-referential Analysis, our goal was to consider the interaction of different attributes and we tried to visualize it in a map over some Animation/Video. With the videos presented in this section we want to discover tendencies in our data.

1.3.1 Type of Weapons used by Terrorists

This time we considered the Type of Weapons used by Terrorists. They are broadly classified into 13 categories:

• Biological
• Chemical
• Radiological
• Nuclear
• Firearms
• Explosives/Bombs/Dynamite
• Fake Weapons
• Incendiary
• Melee
• Vehicle
• Sabotage Equipment
• Other
• Unknown

1.3.2 Weapons Overview

Here, 'Biological','Chemical','Radiological','Nuclear' are considered as one group namely 'Biological/Chemical'.

As we can see, the weapons that are used mainly are:

• Explosives/Bombs/Dynamite
• Firearms
• Incendiary
• Melee

1.3.3 Distribution of Weapons over different types

A different view for the above analysis.

1.3.4 Weapon types associated with different regions

Here, Weapons types are considered over different regions of the world. We wanted to see if a certain weapon is used extensively in a certain region, or if the certain weapon is easily available in a region and so on.

As we can see, weapons associated with major regions are:

• Australasia & Oceania :- Incendiary/Firearms
• Central America & Caribbean :- Melee and Incendiary/Firearms
• North America :- Incendiary/Firearms
• South America :- Incendiary/Firearms and Explosives/Bombs/Dynamite
• East Asia :- Melee and Explosives/Bombs/Dynamite
• Southeast Asia :- Incendiary/Firearms and Explosives/Bombs/Dynamite
• South Asia :- Incendiary/Firearms and Explosives/Bombs/Dynamite
• Central Asia :- Incendiary/Firearms and Explosives/Bombs/Dynamite
• Western Europe :- Mixed
• Eastern Europe :- Explosives/Bombs/Dynamite
• Middle East & North Africa :- Incendiary/Firearms and Explosives/Bombs/Dynamite
• Sub‐Saharan Africa :- Incendiary/Firearms

2 - Conclusion and future steps

In this point of our work, we concluded the main exploration of the data. We described terrorism (week 2), terrorism targets (week 3) and terrorism perpetrators (week 4 and 5). Besides, we conducted some correlations, analysis of missing values and clustering analysis and computed our final dataset with the relevant data. In the future work, we aim to start developing predictive tasks.

3 - Weekly Presentation

https://docs.google.com/presentation/d/1CDnPj8zKSp1GiJn646kjHH8HQbDIOE2XdKXaWfkxg4U/edit#slide=id.g35f391192_00

4 - Perceived Feedback

• Include frequencies in the terrorist success percentages
• Maybe use the "unknown" label for the groups when analyzing certain types of attacks
• Include frequencies in the plots for the clustering and suicides
• Try to complete the "unknown" labels for the terrorist groups, given the place where that attack happened.