What do you think of when you hear phrases like predictive analytics, data mining or machine learning? For many people these terms sound suspiciously like ‘statistics on steroids’ and unfortunately, even in the more data-centric and numerate industries, that isn’t likely to elicit the most enthusiastic of responses. To be fair that’s hardly surprising, as I’m prepared to bet that you won’t have encountered that many people with especially fond memories of undergraduate stats modules (unless of course you teach a undergraduate stats module in which case this might be news to you).
So let’s ask a slightly different question – can you think of examples of applications of predictive analytics? You may be aware of smart systems in the commercial world that aim to automatically present the ‘next best offer’ to subscribers, uncover useful customer segments from historical data, forecast the weekly demand for key products or detect insurance fraud in real time. You probably will already know that these kinds of initiatives are not going to be driven via traditional BI/MI platforms because ultimately they rely on predictive modelling algorithms or techniques like cluster analysis and anomaly detection that uncover patterns in the historical data in order to generate the necessary critical insights that make them so compelling.
If that fills you with dread, you may be interested to know that increasingly these same algorithms are able to generate these critical insights in the form of recommendations, likelihood scores and estimates via automated processes. In which case, if you are thinking about moving beyond traditional BI and making your first foray into the predictive analytics sphere, this may well be welcome news.
In truth although we may never completely get away from the fact that successful predictive analytical applications are ultimately underpinned by sophisticated algorithms (which in-turn are underpinned by some pretty complex maths), you can at least stop worrying about becoming an expert in complex modelling techniques and instead focus on some of the other aspects of developing successful predictive analytics applications. In fact if you get these other aspects right, the whole business of building your first models will become a lot more straightforward. In my next blog post I'll talk about how to get the basics of a predictive analytics project right.
