Predictive Analytics

Predictive Analytics

Predictive Analytics is often considered the apex of data analytics.  For good reasons, it is also often the most complicated and poorly understood activities in a data analytics service offering. At it’s heart it seems simple: using historical data, build a mathematical model that can accurately forecast future activity.

NanoLogic knows that it is rarely a simple process.  Sometimes the historical data isn’t clean enough to support predictive modeling.  Sometimes the method used for collecting the data has changed severely over time and the “historical” value of what data is available may be suspect.  Sometimes, the data is tainted by “selection bias” or the model is “overfitted” and the accuracy of the prediction suffer as a result.

There also literally dozens of competing mathematical techniques: neural nets, Naive Bayes classifiers, linear or logarithmic regressions, decision trees, ensemble models to name a few.  Which are appropriate for use in your data analysis?  How do you train your data?  How much of your data should be set aside for training and how do you sample your data?

Finally, once you do have a predictive model with a decent level of accuracy, how do you keep that model up to date and how do employ it in your operations?  We’ve helped companies build real-time predictive analytics into their email and mail-based marketing campaigns, into their social media monitoring efforts and into their phone systems.

Ask NanoLogic how we can move your company to the forefront of predictive analytics.