Machine learning

In the past two decades, terrorism activities from across the globe have led to hundreds of thousands of deaths and left millions homeless. It has dashed the economic, political, and humanitarian concerns altogether.  After touching record numbers in 2014, terrorism activities worldwide have been on the decline for the past few years after Islamic State and Boko Haram faced major defeats. But the rising dominance of the Taliban in Afghanistan and their immediate take over of the capital, Kabul, has brought the conversation back to the desk. Terrorism is the biggest enemy for world leaders and their motto is to completely eradicate such adverse activities. In order to streamline the motive, governments are also looking for disruptive ways to combat violence. As a result, researchers have come up with a trailblazing initiative. Interpretable machine learning predicts terrorism based on previous data. It signals the areas where terrorism can emerge, signaling even the slightest possibility of violence. 

Governments across the globe are threatened by the multifaced nature of terrorism including their ideologies, motives, actors, and objectives. Over two decades back when a myriad of terrorist attacks collapsed the World Trade Center in New York and killed at least 3,000 people, the world came to the realization that militant groups could start becoming a big threat to the social structure of a country. Starting then, many researchers have engaged in developing artificial intelligence-based models that could predict terrorism activities or sense adverse thoughts in people’s minds that could make them take up violence. Following this quantitative research was developed and used to map the spots of possible terror attacks. 

As a result, researchers came to a solution where machine learning predicts terrorism and detects them at its best. Previously, a study conducted unraveled that machine learning approaches can predict the region and country of terrorist attacks. The machine learning algorithm used data collected by the Global Terrorism Database (GTD) between 1970 and 2017 and detected possible terrorism targets with 82% accuracy. While long-term terrorist attacks are easy to find, short-term plannings and unprecedented attacks come under severe scrutiny. Even the machine learning algorithm was unable to predict it with good accuracy so far. Machine learning models can identify terrorism activities easily, but it also raises a statistical, practical, and recursive problem. When it comes to identifying unplanned attacks, the machine learning algorithm is ineffective, risky, and inappropriate. It also flags a correct alarm for every 100,000 false positives. Recent research led by Dr. Andre Python from the Center of Data Science at Zhejiang University investigates on machine learning algorithms that have aimed at describing regional cases of terrorist acts that provide reliable and accurate short-term predictions at the local level.

Unravelling Machine Learning Algorithm that Predicts Short-term Attacks

A study published in Science Advances examines how machine learning algorithms are capable of predicting fine spatio temporal scale in collision with the occurrence of terrorist attacks. The research team led by Dr. Andre Python has covered all regions affected by terrorism and militant groups over a long period of time. They took data from 2002 to 2016 that covers over 26,551 grid cells at 50kmx50km in over 795 weeks. Combining predictive models with machine learning algorithms helps them clearly visualize the probability of the occurrence of terrorism.