What is Predictive Analytics?

Predictive analytics is a field of business intelligence, statistics and machine learning that has emerged from academia over the last decade or two.  It’s become the state-of-the-art method for analysing business data in real-time or otherwise for predictability of commercial endgames.

Predictive analytics is what drives Amazon.com’s ‘You might also like this’ suggestions. A telecom might use it to determine which customers need a special deal to stop them from switching to another company. An email marketer might filter their list to avoid annoying off-target customers.

Predictive analytics is any method that allows you to predict what your customers are going to do. Just by tracking your customers’ purchases and noting what other items customers who bought item X bought is a simple form of predictive analytics. But in it’s truest sense predictive analytics is used to discover patterns beyond the non-obvious.

Using simple statistics methods like MLR (multiple linear regression) predictive analytics can uncover how non-obvious correlations might exist between your customers’ purchase history or their demographics and their future purchases as determined by historical data. Not only can MLR discover the obvious two or three correlations (like lots of purchases mean lots more) but that your other ten variables also carry additional predictability (like from the customer’s particular combination of demographics).

The only limitation for simple methods like MLR are that they can only handle linear correlations. For example, what if some variables combine in strange ways such as: a shopper is likely to buy more food if they’re from a poor demographic but purchase luxury items if they are in a wealthy demographic. These sort of complex intertwined correlations can only be discovered by machine learning algorithms like artificial neural networks (ANN), decision trees or support vector machines (SVM). These methods, that require genuine expertise, can add 10-30% more predicability over and above what MLR can provide. And as I always say, that means 10-30% more sales.

To contact us click here.

Posted in Machine learning, Multiple linear regression, Predictive analytics | Leave a comment

What is Machine Learning?

Machine Learning (ML) is a class of pattern finding methods that lies at the intersection of statistics and Artificial Intelligence (AI) and, algorithm-wise, represents the state-of-the-art of predictive analytics. ML methods include Artificial Neural Networks (ANN), Decision Trees (DT), Clustering algorithms and Support Vector Machines (SVM) to name a few.

These methods always (when done correctly) generate better predictions, typically 10-30% better, than the base-line MLR (Multiple Linear Regression) simple statistics method.

These Machine Learning methods work by fitting far more complex non-linear relationships between your predictor variables and your target prediction output. So, for example, ANN (artificial neural networks) fit your historical data to your desired prediction by adjusting the strength of thousands of connections (synapses) in neuron-like algorithmic structures during ‘training’ on your historical data. Once trained, the ANN ML system can predict future events and will have very similar accuracy to that gained during training if verified during training by a test subset of data.

The upshot is that even non-linear relationships can be modeled so your prediction is more accurate than the base-line MLR method which is linear by nature.

The take home message is that, machine learning, when done correctly, can enable you to use the full information content in your customer data to it’s 100% potential. Although one can argue about which ML method is the best, once you settle on the best one for your data you can be confident that there is almost nothing else you could do to get more bottom-line information out of your data than by using ML.

To contact us click here.

Posted in Artificial Neural Networks (ANN), Machine learning, Prediction verification, Predictive analytics | Leave a comment

How Does MLR, the Simplest Predictive Analytics Method, Work?

MLR or Multiple Linear Regression, works by optimally combining all the predictor variables into a simple ‘weighted linear combination’. All that means is that we say our prediction (eg the chance a customer is going to switch away from you) is equal to a sum of all the predictors (purchase amounts, purchase dates, demographics etc) with an important weighting factor in front of each predictor. These weighting factors, which may be negative or positive, depend on the scaling of each variable (it’s range of minimum to maximum values) and, critically, how well it correlates with the predictor using historical data.

This is the base-line method of predictive analytics and will typically tell you how predictive your data is with some caveats. And it will usually be within 10-30% of the best prediction you will ever devise using more advanced methods.

MLR is what we do on day 1 after cleaning up your data. That way we don’t waste your time and budget with complex analyses if there is no predictivity in your data.

After that it’s time to bring out the big-guns of predicative analysis, the logistic analysis and the machine learning algorithms including artificial neural networks and the like. they’ll get you an extra 10-30% predictive accuracy which, to get to a bottom-line, can translate to 10-30% improved profits on top of whatever gains MLR got you.

To contact us click here.

Posted in Data clean up, Multiple linear regression, Predictive analytics | Leave a comment

How Do We Start Using Predictive Analytics in Our Company?

At Antara PA we recommend, regardless of your company size, that you should talk to a predictive analytics specialist like us before purchasing advanced analytics and visualization tools or choosing an implementer. We can show you how predictive YOUR data is for your bottom-lines, obligation free. Then, if you chose us, we can do a full analysis and an implementation recommendation, whether to be done by us or otherwise. Whatever way you go, you don’t waste your budget with us.

The alternative is to purchase sophisticated tools and try and nut-it-out yourself, and/or sign-up with an implementation consultancy where the predictive analytics expertise may reside in a ‘module’ expert who may not even be a data scientist.

The upshot is, there is nothing to lose, and a lot to gain, in doing a quick analysis with a consultancy like us where predictive analytics is all we do.

To contact us click here.

Posted in Predictive analytics | Leave a comment

Low-hanging Fruit of Predictive Analytics . . and Then the Rest

Most of the predictive power of predictive analytics comes after the all-important data clean-up and the application of a 1st-year uni stats method: Multiple Linear Regression (MLR). We’ll talk about what that is in another post, but for now, just appreciate that MLR is a really straight-forward technique that allows us to see how ‘smooth-changing’, ‘one-directional’ variables can combine to predict your desired output variable.

And most of the time that MLR method will get you to within about 10-30% of the best prediction. It means that at Antara PA we can get you a quick answer on whether your data is predictive.

After that, you can decide whether we apply the next level of tools in the armoury: machine learning methods like Artificial Neural Networks (ANN), Decision Trees (DT) and Support Vector Machines (SVM). These need a real analytics expertise and will improve your prediction by 10-30%.

Take home message: we can rapidly tell you whether your data is predictive. But usually you’ll want to go the next step. If your prediction is really working, why throw away 10-30% of your customers?

To contact us click here.

Posted in Data clean up, low-hanging fruit, Machine learning, Multiple linear regression, Predictive analytics, Uncategorized | Leave a comment

Why is Predictive Analytics Better Than Just Watching the Key Variables?

Sometimes it’s tempting to wonder whether all this predictive analytics is worth it, and instead, just watch your key variables. But here’s where the fallacy lays: it’s the subtle extra information in the other variables which can typically double your predictability.

Now, by trial and error, you could try and manually search for these subtle additional predictors that together add significant predictability. But, firstly, that’s the point of predictive analytics, and machine learning in particular, once configured it does it for you automatically and optimally. Secondly, and perhaps even more importantly, machine learning will continue to learn even as your data changes.

To contact us click here.

Posted in Predictive analytics | Leave a comment

Do I Need to Understand Everything About Predictive Analytics?

The short answer is no, thank goodness.

However, if you purchase expensive visualization and analytics tools you’ll probably want to put the effort in to get your money’s worth. So in that case you will want to become somewhat of an analytics expert.

But, if instead, you rely on a consultancy like ours which can do all the key leg-work using open source tools and at a fair charge out rate, we can inexpensively take you through the predictive analytics learning curve at the level you want to be involved. You might be happy with us showing you that 76% of your predictivity comes from these 3 variables and an additional 24% from another 17, or that 81% comes from the simple MLR analysis and the remaining 19% comes from the sophisticated ANN algorithm. That might be sufficient to then let us implement a solution without your needing to become a data scientist.

And of course we can readily manage the translation of our open-source investigations into the language of your preferred legacy implementation at the end if necessary, once we’ve determined where your predictive advantage lies.

To contact us click here.

Posted in Legacy, Open source, Predictive analytics | Leave a comment