We can predict the time frame of terror attacks: Expert

It may seem strange if one says that terror attacks can be predicted. V S Subrahmanian, professor of computer science and a Fellow of the Association for the Advancement of Artificial Intelligence, says it can be.

Subrahmanian, co-author of ‘Computational analysis of terrorist groups: Lashkar-e-Taiba and Indian Mujahideen: computational analysis and public policy,’ currently heads the Centre for Digital International Government at University of Maryland. In an e-mail interview with Shemin Joy of Deccan Herald, he speaks about data mining, predictions on Indian Mujahideen and how data-mining helps authorities to build a policy framework.

Excerpts:

Can you explain the science behind predicting the conditions that can give a hint about a terror attack through data-mining?

We collected data on 770 variables that may have been linked to terror attacks. These variables include social, economic, political, religious, and internal operations of the groups being modelled. Attacks are triggered by a combination of motives like desire to attack, revenge for arrests or killing of operatives, desire to show recruits that they are strong, technical ability to carry out attacks, which is a given in the case of IM because of ISI support and opportunity.

Most IM attacks in India have been in public places like markets, so opportunity is present. We can predict time-frames when attacks may occur by using data mining on historical data to learn when IM has the motive, means, and opportunity to execute attacks – and our 770 variables are geared towards capturing past data on these.

Our predictive capabilities provide less granular predictions than we would like. We cannot, at this stage, predict that they will attack a particular location at a particular time. We can predict that they will attack certain types of targets, for example security installations, during certain time frames, usually a 3-month window. Current predictive models such as ours can be used as inputs by law enforcement agencies to determine when to increase intelligence collection and when to increase security precautions.

How does data mining help predict Indian Mujahideen attacks?

Data mining broadly shows that the IM carries out attacks (i) a few months after India-Pakistan diplomatic relations start warming, (ii) a few months after IM organises a meeting of its leaders, (iii) a few months after there is increased IM public communications, e.g. reports of anger about certain events, (iv) a few months after arrests of IM personnel and reports of increased cooperation between IM and other terror groups like SIMI or LeT.

Our models currently predict that IM will carry out attacks in the January-May 2014 time-frame. Most likely, these attacks will be in public places such as markets. In the past, IM attacks have focused on Varanasi, Mumbai, Delhi, Hyderabad, and Bangalore, but we cannot say which of these locations will be targeted.

How would data mining and analysis help build a policy to rein in terror outfit and prevent future attacks?

Our data-mining engine generates ‘rules’ like ‘when some condition C is true, then IM will carry out attacks in the next few months.’ When C is true, an attack occurs with high probability, when C is false, they occur with low probability. These rules can be used both to predict attacks and in some circumstances, pre-empt them.

Another rule indicates the high likelihood of attacks when IM is receiving substantial assistance from other terrorist groups. Disrupting these links could prevent IM attacks. These last points require building detailed methods for travel intelligence. Where are people going when they leave India or prior to entering India? For instance, a number of IM and LeT personnel go from India to a neighbouring country and travel on a fake ISI issued passport from there on.

To monitor this, India needs to help friendly neighbours improve their border security systems, enter into travel information sharing agreements with them, and build ‘travel maps’ of people’s movements.

Besides terrorism, India has many internal security problems. Do you think data mining could help in tackling Maoists, as well?

Studying the Maoists is one of the next priorities for my research group. Big data could also be useful in examining the behaviour of organized crime networks or studying trends in general criminal activity.

Social media has become a major driver in opinion making. India is going to have a general election and there is a big talk about politicians using technology to reach out to voters. What do you think of it?

Analysis from social media is highly predictive of trends in voter sentiment. It could play a key role in India where social media penetration is amongst the highest in the world. We have used it in the past to predict outcomes of elections in France, Italy, and Pakistan.

How suited is data mining for India and how far India has progressed? Has policy makers at any point of time approached you seeking guidance or help?

 I have been approached by policy makers in many countries, including India. India's software industry is very well aware and competent in data mining methods and their applications in security and counter-terrorism. I am sure they would do an excellent job in helping India fight terrorism. In addition, one of my co-authors on the book on IM is RK Raghavan who served as the CBI Director.