The conference covers a variety of predictive analytics topics from a practictioner's point of view. This means that techniques and market niches that can improve your business are on focus here.
The overall impression is that it is a very well organized conference, very useful for networking since it puts together many very good data analysis practictioners, very interesting to get new ideas or to catch what the hype in data analysis is at the moment.
Not all talks were very detailed about the methods and techniques that they use or about the problems that they face. Some talks were very general about the market and the amount of data around us. Nevertheless I have seen some interesting presentations where I got some nice ideas of where my business can go in the future.
The organization was impeccable. Talks were on time, media were working. During the pauses and lunches it was easy to meet and talk to other people.
Particularly nice were the "birds of a feather" lunches where tables had a designed topic (like "Time Series Analysis" or "Data preparation"). This helped you to join the right crowd without having to spot the people yourself.
There were 2 parallel sessions and only for the keynote speakers the 2 sessions were reunified into 1. I did not listen to all the presentations of course. I report here only my impressions about interesting points and trends that I have heard.
Data from Social Media. A few talks (Case Study: "1800 Flowers", "Younoodle", "A Leading North American Telecom") were insisting about using data from social media to detect bad people, mainly fraudolent customers. It seems that people who are fiend with good people have a high probability of being good themselves. However, people who are fiends with bad people does not mean that they are bad themselves. Another discriminating factor lies on the number of connections. Bad users have a small number of connections. The explanation that I found is that fraudolent customers might build up fake profiles on social media. Usually fake profiles will not have abig number of connections, because they are fake and because they have been created only recently.
Cloud Computing. Another set of talks (Case Study :"Visa", "Google", "In-Database vs. In-cloud Analytics") concentrated on the big size of the data analysis problems and the need of an ever-growing machine power. Data analysis models are now easy to build in a large number. While parallel computing can help with their execution, it was debated if the cloud computing was ever going to outdate current databases.
Text Mining. A third line of talks was about text mining (SAS, and others).
My preferred talk. My preferred talk was definitely the keynote talk from Kim Larsen, "Response Modeling is the wrong modeling: Maximize impact with Net Lift modeling". The problem was clearly explained and the methods were described in details, not only technically but also in terms of what has helped and what not.
All in all it was a great opportunity to network, since you could find there a lot of experienced practitioners in the field of data analysis.