Predictive Software — EmTech 2009

At 5 in the morning, the smart phone is beeping and vibrating its lungs out.
A mail from the agency is usually not anticipated at this time unless you're
working on something important, something classified and of extremely high
value. Being a forensics and security expert, a call from the anti-terrorism
squad being run by the government isn't surprising anymore for john. After
reaching their HQ, he is briefed by a senior officer. This time it's an email
message containing what seems like codes compiled in English language and
numerals from a source that is bothering the squad. With more than a decade of
experience in data mining and analysis, John is now supposed to make sense out
of the mail containing this amalgam of words as the mail poses a potential
threat to the state.

This might sound like the storyline of the next version of Die Hard or any
other out of the world insanely big budgeted Hollywood flick. But at EmTech 2009
held in New Delhi, Srikanth S. Sampara, chief executive of Tuple Technologies
gave something similar as an example where the concept of predictive software
can come in handy. When using an appropriate model and technique, predictive
software can also help in pointing out possible fraudulent cases, such as in the
case of insurance claims, said Sampara when asked for a day to day world
implementation. But in a case like this, reliability and success rate play a
very important role as a fraudulent application being passed as genuine one for
sure would bring many insurance companies on their knees begging for bailouts.

Computer scientists are taking advantage of vast databases and sophisticated
modeling techniques to create algorithms that help in making correct predictions
of changes in financial markets, business opportunities, traffic patterns and
even international conflicts. Dr. Lipika Dey who is a senior scientist at the
Innovation Labs of Tata Consultancy Services divided data processing into two
categories, structured and unstructured both requiring different modeling
techniques and levels of complexity. Different models are being tried and tested
in order to extract actionable intelligence from enterprise data.

Dr. Lokendra Shastri who is the General Manager at SET Labs, Infosys said
that Languages can be ambiguous. Specially in the web age, where short forms and
new words are being formed each day, deriving inferences has become more
complex. He gave a few examples of simple sentences which could have more than
one meaning in different situations. Dr. Shastri has worked on areas of scalable
, parallel models of knowledge representation and inference and showed how these
play a significant role in data mining and analysis techniques.

Could a predictive software solution warn us about the ongoing financial
crisis? The answer to that was probably 'no.' Adding on to his answer, Suresh
Satyamurthy, country director of Autonomy Corporation said predictions and
enough warnings couldn't be made by thousands of Wallstreet professionals, plus
completely relying on such solutions is not the best thing to do when you have
so much at stake. Maybe in future we would have solutions so intelligent that
will help us in avoiding such situations.