Predictive Analytics: Cross-Industry Challenges Interview with Vijay Desai, Ph.D., Principal Scientist, SAS

To benefit from Predictive Analytics, there are potential pitfalls to avoid along the way. In this interview with Vijay Desai, Ph.D., Principal Scientist, SAS, Dr. Desai shares his ideas and solutions as they compare across multiple industry segments.


Bill Cullifer: I am on the phone with Dr. Vijay Desai, Principal Scientist for SAS, S-A-S. Dr. Desai, good afternoon and thanks for agreeing to the call.

Vijay Desai: Thank you for having me.

Bill Cullifer: You bet. You recently participated in a panel discussion entitled Cross Industry Challenges and Solutions regarding predictive analytics and the theme included hurdles to jump and barriers to avoid when considering benefiting from predictive analytics. Can you summarize that discussion and what you walked away with?

Vijay Desai: Sure. Some of the things that we discussed were #1 analytics is becoming more and more useful in different industries and so as we go from industries where analytics has been used for a long period of time, in new industries there are some challenges. Some of these challenges include first convincing the users that predictive analytics can be of help, that predictive analytics is not going to take away their jobs, but by increasing their productivity it’s going to make them more valuable and help them generate more value and then even the data is becoming more and more available in different industries, you do not have what I would call targets. So, for example, in the case of fraud, you… target would mean identify the accounts that are good or fraudulent accounts. So, when you do not have targets available, it makes your job more difficult. So, in credit card fraud, targets are easily available because if your credit card gets stolen, you will report to the credit card company pretty soon that it has been stolen and so the credit card company can tag your account as being in a fraudulent state whereas in other industries where we are trying to use predictive analytics now say for example tax under filing, the tax under filers are not going to report that, “Hey, I am under filing my taxes this year.” So, you do not have that target information. Also, to take things further if you are… I have also worked in the area of network intrusion detection where you are monitoring a computer network, you don’t know which particular activity is malicious and you know which one is normal. So, again, you do not know the targets. After the fact, using forensics, you can identify some of the activities as being malicious or not, but you still are not sure you know whether you caught all the malicious activity or not and then going even one step further for bio surveillance; fortunately, we do not have any examples of bioterrorist attacks and so when you are building models, when you are trying to use predictive analytics for something like bioterrorism or bio surveillance, it is a much more of a challenging problem.

Bill Cullifer: Thank you. I certainly appreciate your perspective and for your time today.

Vijay Desai: Most welcome.

Leave a Reply

Your email address will not be published. Required fields are marked *