Title: Using Easymine-An automatic intelligent model building system to develop products sequential recommendation system

Biography:

Morgan C. Wang is the funding Director of Data Mining Program (funded in 1999) and Professor of Statistics at the University of Central Florida (UCF). He is also an affiliated faculty with the School of Computer Sciences and College of Business Administration at UCF. He couched student teams to win the 2011 and 2012 SAS Data Mining Shootout Contest. He won the best conference award in the First Annual Conference on Engineering and Technology Innovation in 2008. He was the first prize-winner in Data Mining Competition of the 11th SIGMOD KDD (the most predigest data mining competition) conference in 2004 and the first prize winner in Data Visualization Contest of SUGI 25 conference in 2000, and was given invited talks on making intelligent decision based on big data analytics for more than eighty times for American Statistical Association, SIGKDD (leading conference in data mining), International Conference on Information Technology, SAS Global Forum, Well Fargo Bank, Republic Bank, Florida Blue, Disney, Kemper Preferred Auto Insurance, HealthFirst, QFOR, and many companies and universities around the world. He is an member of Ad Hoc Big Data Advisory Committee for the President of American Statistical Association (ASA).

Abstract

“EasyMind” is an automatic intelligent model building system.  This system has five components: data exploration component, data preparation component, model building/validation/selection component, result automatic generation/data scoring component, and model understanding component. 

This system has data preparation component that can fix data problems such as missing values, skewness, and high cardinality.  In addition, this system has modeling component that can fine tune the model parameters to build a “better” model.  Currently, it supports neural network, decision trees, gradient boosting, rand forest and many regression algorithms.  After the optimal model selected, the user can further test the model performance or use the selected model to score new data.  This system also attempts to open the black box to allow the user to see some insight of the modeling results such as interaction among predictors, important predictors, how to alter predictors to change the predicted values.

This system has successful used on developing sequential production recommendation system for a big bank in China. Instead of recommending one product to its’ potential customer, this system can select an array of products and recommend these products to potential customers.  Experiment results have shown that an average of six-fold increasing.