YOU ChangLi CHEN Jun ZHANG Li

Abstract： For terrorist activities， according to the data in GTD， Pearson correlation coefficient analysis was conducted on the characteristics related to terrorist attacks. Five indexes including the location of terrorist attacks， whether the attacks are successfully launched， whether the attacks are suicide attacks， the type of targets and the nationality of targets are selected as the correlated indexes of study. Also， the correlated model between terrorist event and terrorist organization is constructed by using naive bayes classifier. After the analysis of the terrorist events that account for 82.6% in all terrorist attacks and that have not been claimed responsibility by any terrorist groups， this paper has speculated out the probability of suspicion of the terrorist groups， improved and amended information on database of terrorist attacks and improved the accuracy of terrorist attack prediction by 5-40%， making the analysis and description of information of terrorist groups more accurate.

Key words： counter-terrorism intelligence， terrorist attack， naive bayes， the probability of suspicion， prediction

1 Introduction

According to the Global Terrorism Database， from 1998 to 2017， there were more than 110，000 terrorist attacks and 279，280 deaths worldwide in the past 19 years[1]. Terrorist attacks have been threatening the peaceful life of mankind. All countries in the world actively cooperate with each other to find solutions to global governance of terrorism and set up efforts to fight against terrorist groups [2]. In view of the existing problems， this paper proposes a naive bayes classifier as the basis to establish the analysis model of the correlation between terrorist event and terrorist organization， to improve the deficiencies of this aspect of research， and to provide information which can be helpful to speculate the real perpetrators of terrorist attacks.

2 The Source and Analysis Data

In this paper， GTD is used as the main data resource， Country Reports on Terrorism and International Terrorism and Counter-terrorism Yearbook are used as auxiliary databases to complete the missing data. The number of global terrorist events shows an increasing trend. During the past 19 years， a total of 113，983 terrorist events occurred， among which 94，094 terrorist events were claimed by no groups， accounting for 82.6% of the global terrorist events. Therefore， it is of great significance to establish an effective correlation analysis model between terrorist attacks and terrorist groups.

3 The Construction of Correlated Model between Terrorist Group and Terrorist Event

In this paper， the correlation model between terrorist group and terrorist event is analyzed by using existing data as a training set. Firstly， select relevant indexes， some groups are selected from all terrorist groups as samples. Then， guided by bayesian priciple， naive bayes classifier is trained. Finally， the classifier is substituted into all terrorist attacks that have not been claimed by groups to speculate the perpetrators.

3.1 The Selection of Model Indexes

According to the research on the literature of Jiwu Yin and Xiaohui Hu ， the features of terrorist groups are actually obvious， relating to the place where the terrorist attack occurs， whether it successfully launches， whether it is a suicide attack and the type of attack. By using SPSS software and Pearson correlation coefficient， this paper analyzes all terrorist attacks launched by confirmed terrorist groups from 1998 to 2015 and obtains the table of correlation between these eight characteristics and terrorist groups. This paper finally selects features that are significantly related to terrorist groups： region， whether the attack is successful， whether the attack is a suicide attack， target type and target nationality.

3.2 The Selection of Terrorist Groups Samples Using Reservoir Sampling Algorithm

Reservoir sampling ensures that the probability of each group being selected is the same. This paper adopts the reservoir sampling algorithm under the equal probability sampling algorithm to select 10 groups from all as samples. Calculated by Python and using reservoir sampling algorithm， finally 10 terrorist groups are selected as the samples. They are： Abu Sayyaf Group （ASG）， Hamas （Islamic Resistance Movement）， Tehrik-i-Taliban Pakistan （TTP）， Sinai Province of the Islamic State， Chechen Rebels， Al-Aqsa Martyrs Brigade， Al-Shabaab， Tripoli Province of the Islamic State， Baloch Liberation Army （BLA）， Boko Haram.

3.3 The Construction of Naive Bayes Classifier Model

With low variance and high bias， naive bayes classifier adopts Attribution Conditional Independence Assumption to assume that all attributions that independently affect the classification result are independent of each other based on known categories[4]. As shown in Formula 1.

According to table 3， it can be concluded that the number of terrorist attacks actually launched but not acknowledged by terrorist groups can be estimated by using the correlated model of terrorist attacks and terrorist groups. After completing these data and comparing the actual number of terrorist attacks with the results of the two predictions， it was found that the prediction error rate of future terrorist attacks could be reduced by 5%-40%. A more accurate prediction can be achieved. Also， the revised database can also provide more accurate data for the research of the development of terrorist groups.

5 Conclusion

Based on the naive bayes classifier， this paper speculates terrorist events of unknown perpetrators， finding that a number of terrorist attacks are the heinous actions of existing terrorist groups. However， the naive bayes classifier is not further improved in this paper， so there may be errors in accuracy. In the future study， the classifier algorithm will be further improved to increase the accuracy of the classifier， so as to achieve better effects.

Reference

[1]GTD [EB / OL]http：/ / www. start. umd. edu / gtd / .

[2]Tang Chao， Meng Xiangyun. Research on Surrounding Terrorism-Related Safety Risk Environment Based on GTD[J]. J Intell. 2017（05）.

[3]Yin Jiwu， Qi Yifang. Three Extremes Thoughts behind the Terrorist Attacks in Europe [J]. Peoples Tribune，2016，（z1）：105.

[4]Wang Xingfu， Du Ting. Improved weighted naive bayes classifier algorithm based on attribute selection[J]. Computer Systems & Applications. 2015（08）.