Suppose you’re trying to build a model to predict respondents, and in your data set, about 3% of the population will respond (target = 1). Without applying any specific analysis techniques, your prediction results will likely be that every record is predicted as a non responder (predicted target = 0), making the prediction result insufficiently informative. This is due to the nature of this kind of information, which we call highly imbalanced data. Read more

Even though we’re already blazing full speed ahead into 2016, it’s always important to take a minute to look back at the past year and recognize the high points that made it special. In addition to being named a Boston Business Journal Pacesetter for the second time, making the Inc. 5000 list of fastest-growing companies for the third time, and receiving IBM’s 2015 Business Intelligence Partner of the Year award, we’ve produced several valuable and popular pieces of thought leadership to enrich the analytics community. Here are the top 5 articles 2015 saw us release. We hope you find them useful as you start your journey into the new year. Read more

Do you ever wonder how Netflix makes recommendations for you? Or how the drug store decides which coupons to offer you when you make a purchase? Behind the scenes they have a data scientist conducting what is called market basket analysis, which searches through vast amounts of purchase history information to find patterns in people’s purchases, web searches, or Netflix viewing preferences. The data mining technique used for market basket analysis is called Association Rules (AR). This is the actual algorithm designed to detect probabilistic if- then statements, such as “If you watched Breaking Bad and House of Cards, then you are also likely to enjoy Mad Men.” Read more