Tapan Patel and I recently caught up with Dean Abbott, President, Abbott Analytics for a Q&A session on Predictive Analytics World that takes place during Data Driven Busines Week March 14-15 in San Francisco.
Question: Any thoughts you would like to share for upcoming Predictive Analytics World 2011 Conference? What role are you playing at the conference?
Answer: I always look forward to PAW because it provides business-centric or solutions-centric view of predictive analytics. I think that is why one sees such an interesting mix of companies and individuals who are expert in predictive analytics alongside those who are new to predictive analytics. I've had several of my own customers go to the conference and all have had very positive experiences there.
Question: What are the top reasons for attendees to sign up for Hands-on Predictive Analytics workshop featuring SAS Enterprise Miner?
Answer: There are several reasons to attend the Hands-On workshop at PAW. First, it exposes attendees to the Predictive Analytics who perhaps have never experienced it before. Predictive analytics contains both art and science. The science part one can read in any number of excellent books on data mining and predictive analytics. However, predictive analytics can't be done effectively as a recipe; it takes experience to recognize problems that can arise during ones analysis, and possible ways to overcome those problems. The workshop will provide a framework for how to think about solving problems with predictive analytics techniques.
Second, predictive analytics doesn't take place in a mathematical vacuum; one uses a software tool to build solutions. The attendees will use SAS Enterprise Miner to solve the problem posed for the day, a premier predictive analytics tool. If an attendee belongs to a company that already has SAS or even Eneterprise Miner already, this course will provide a head-start for them to know how to bring predictive analytics to their workplace. If they are new to SAS or Enterprise Miner, they will be introduced to a premier predictive anaytics software tool. It is essential that the tool we use for the workshop be powerful enough to perform effective analyses, but also easy enough to learn that one can be proficient in a day. The best response I've received after a hands-on workshop was one attendee stating "I can do this!"
Question: What do you think are some of the common challenges customers face before embarking on a predictive analytics project? How will the full-day workshop address some of the challenges?
Answer: The most common issue I see with those new to predictive analytics is a lack of appreciation for the importance of tying the analysis to a business framework. The most effective predictive analytics solutions tie the target variable to be predicted to a business decision that will improve profitability, efficiency, customer satisfaction, or some other measure. But then, the way the final predictive model is selected is best done in the context of this business decision: pick the model that most effectively improves the decision. The workshop begins and ends with this perspective.
The second most common challenge is gaining an appreciation for how clean and consistent data needs to be for predictive models to behave in the way we intend them to. Data cleanliness from a predictive analytics perspective can be quite different from data cleanliness from a DBA's perspective. We spend time during the workshop considering how we would like to represent the data so we can overcome these potential problems.
Third, many are intimidated by algorithms like neural networks, especially how one can build them and what they mean. In the workshop we show how easy it can be to build predictive models, some common-sense settings one can use to build them, and how one can automate the process so building models is not so time consuming.
Question: Based on your customer engagements, what are some of the hot new areas in predictive analytics offering the most potential in next 3 to 5 years?
Answer: The biggest growth area I see is in text mining or text analytics. Unstructured data has long been recognized as containing a wealth of information that would be useful in making decisions operationally, but most companies found the hurdles to gaining insight into the text too high or expensive. This is changing rapidly with the onset of easy-to-use software tools that can turn the unstructured data into structured fields to augment already-existing structured data. I see text mining becoming more automated and mainstream.
I believe a second area of welcomed growth in the industry is gaining a new (and I would say better) perspective on the problems we should be solving with predictive analytics. Net Lift and Uplift modeling are examples of this perspective. They do not introduce a new algorithm, necessarily, but rather emphasize selecting and creating a target variable that better describes customer behavior. I see the same phenomenon in risk modeling. The more tightly the target variable is tied to an operational decision, the more effective models can be in improving those decisions.
Special thanks to both Tapan & Dean for taking time to talk to us about Predictive Analytics World. Stay tuned for more pre-event interviews regarding the upcoming March 14-15 conference! Please let me know what you want to hear more about prior to Predictive Analytics World.