AI was able to predict terrorist attacks a week ahead using only open data

AI was able to predict terrorist attacks a week ahead using only open data
AI was able to predict terrorist attacks a week ahead using only open data

In recent decades, people in most countries have come to terms with the persistent risk of terrorist attacks. For law enforcement agencies, preventing such tragedies is a direct responsibility and a severe headache. Fortunately, scientists from all over the world have long been studying the phenomenon of terrorism and creating predictive models. And a recent attempt to connect artificial intelligence to this task has yielded impressive results.

AI was able to predict terrorist attacks a week ahead using only open data

An international group of scientists based at Zhejiang University (China) has created an expanded and in-depth methodology for analyzing and predicting terrorist activity. Unlike classical models, it relies on the history of terrorist attacks to a lesser extent. Two dozen structural parameters (which practically do not change over time or do it linearly), such as GDP per capita, the development of state institutions, and the like, have a significant weight in it. The model took into account 14 more procedural characteristics (variables), among which there were already previous acts of terrorism, as well as such unobvious parameters as changes in the illumination of streets in cities visible from space.

The researchers took all information for analysis strictly from open sources and did not use intelligence or other assistance from law enforcement agencies. This approach is necessary, since within the framework of scientific work, the possibility of creating a predictive model accessible to any government and commercial structures was studied. The results of these studies and the work of the resulting algorithm are presented in an article that was published in the journal Science Advances of the American Association for the Advancement of Science (AAAS).

Artificial intelligence was "set" on the space-time matrix. It consisted of two groups of cells - territorial and temporary. The first were 26,551 pieces, the entire inhabited territory of the Earth, divided into squares with a side of 50 kilometers. For each such sector, 795 time cells with a duration of seven days were taken into account (that is, a little more than 21.5 million pieces in total). The last number is no coincidence: this is the period for which data was collected for analysis - from 2002 to 2016.

In other words, the scientists provided the algorithm with data on when and where the attacks took place earlier, as well as a bunch of additional social and economic information. Artificial intelligence, in turn, tried to predict terrorist activity. Since the above time frames for people are long history, they know whether terrorist attacks took place in them or not and where they took place (of course, no one gave out this information to AI).

The results were promising: the average accuracy for the next time slot varied from 0.81 to 0.97 (one is a perfect prediction, zero is a failure, half is an ordinary fortune-telling), depending on the estimation method. Forecasting the severity of the terrorist attacks turned out to be somewhat more complicated, especially if the regions were analyzed where terrorist activity is rare or absent.

However, in any case, the new model using AI turned out to be radically more accurate than all previously created ones.Nevertheless, the authors of the work do not yet consider their development ready for implementation in practice. It requires improvements, increased accuracy and additional verification, since only the basic metrics turned out to be impressive, and if you go into details, the algorithm is still often wrong.

A separate section of the scientific article is devoted to the inaccuracies of the term "terrorist act" itself. In the international legal and political field, there are very different views and sometimes entire countries are registered as terrorists. So for scientific purposes - for the purity of the experiment, so to speak - scientists have designated the acts of terrorism as violence produced by non-state actors in world politics. That is, organizations like drug cartels, militarized radical religious groups and criminal groups.

Popular by topic