Rune stones representing SPSS Modeler 17 prediction

IBM SPSS Modeler 17 and Statistics 23 were officially released at the beginning of March 2015. There are several important changes in licensing structure and system infrastructure, as well as many innovative new functionality enhancements. This article will briefly introduce the enhancements added to Modeler 17 from a practical perspective. Something worth noting as you begin exploring this most recent round of upgrades is that Analytics Server will be heavily leveraged in a lot of these new features.

About Ironside’s Customer Analytics Practice

Ironside’s Customer Analytics practice strives to use predictive modeling to propel businesses forward and upward by delivering information on what will happen in the future or what their customers will do based on certain actions or offers to help decision makers act more quickly and with the highest degree of certainty possible.

Customer Analytics and our practice are not limited to traditional B2C analytics. In fact, our strategies can be broadly applied to a wide range of industries, such as predicting response rate and campaign optimization, predicting patient volume and optimizing staffing scheduling, predicting violent crimes and optimizing the deployment of police forces, fraud detection, and a variety of other scenarios.

This article will illustrate use cases for the new features of SPSS Modeler 17 using different industries as reference points. Through this method of presentation, we’ll give you a practical understanding of what these new features mean from an implementation standpoint and how they can help you further optimize and leverage your predictive analytics infrastructure.

Integration with Prescriptive Analytics

Products Involved: SPSS Modeler 17, IBM Decision Optimization

In their daily lives, people often face this kind of scenario: you are given some type of resource (e.g. funding, staff, a machine) and need to achieve certain goals using it (e.g. minimizing cost, maximizing profits). You also usually have some constraints confining you (e.g. the spending cannot exceed a certain amount, the number of working stations is predetermined, or the production needs to satisfy a certain level of demand). Because of this, you need to make decisions about how to best allocate the resources you have to meet your goals without exceeding your constraints. Raw predictive analytics can give you some expectations to work with, but sometimes you’ll need help to know exactly what these results mean and what path you should take because of them. This is where prescriptive analytics come in.

For example, a hospital’s specialty unit has a certain number of nurses. The incoming patient volume and discharging volume fluctuate hour by hour and day by day. If we can predict the hourly expected workload (incoming and discharging volume) and its variability (standard deviation), how should we then schedule the nurses hour by hour so that we can utilize the smallest number of nursing staff while meeting a certain level of hourly demand?

Businesses with advanced analytics capabilities often face this problem of translating what will happen into what action they should take. SPSS Modeler 17 with IBM Decision Optimization can easily help with that translation. This integrated solution factors in your business goals (minimizing cost, maximizing profit, satisfying at least 90% of the demand, etc) and resource constraints with built-in algorithms designed to model based on the variability of predicted results (predicted demands, sales, etc) to give you the best decision options.

When applied to the hospital case referenced earlier, IBM SPSS Modeler 17 would feed the predicted expected patient volume and variability into the prescriptive model and then allow you to specify your constraints, such as how many nurses in total are working for the unit, when the different shifts begin, how long shifts are, and so forth. Decision Optimization would then decide the minimum number of nurses that need to be on shift each hour according to the predicted patient volume in order to meet the goal you specified (e.g. the staff should be enough with a 90% confidence).

New Algorithms to Model Big Data

TCM

Products Involved: SPSS Modeler 17, Analytic Server

Suppose you are operating a retail store selling thousands of SKUs. You may want to find out which SKU’s sales are related to other SKU sales. Are sales for one of your products driving another product’s sales numbers? Are people in one area following the fashion trend in another area by replicating that area’s purchasing pattern? These are just some of the questions that a decision maker at the head of a retail organization would want to answer. Another potential line of inquiry could be geared around knowing what outside factors are driving SKU sales. How is the economy driving home furnishing purchases? What effect do temperatures or changes in a town’s population demographic have?

With TCM, you can feed in thousands of time series and discover the causal relationship among them, using each time series as a target but also as an input for other time series.

 

Two-Step AS Clustering

The new two-step AS clustering node is different from the old two-step clustering node in that it employs a MapReduce-based automated distributive clustering framework so that it’s more suited to big data clustering than its predecessor. This node runs on Analytic Server only.

 

Geospatial Analysis

Products Involved: SPSS Modeler 17, Analytic Server

One of the biggest functionality enhancements included in SPSS Modeler 17 is its ability to process, analyze, and present geospatial data. There are five nodes involved in geospatial analysis that allow users to read, process, and build supervised or unsupervised models and present their results graphically. The nodes contributing to this functionality are listed below:

a. Geospatial Data Source Node

b. Reproject Field Operation Node

c.  Geospatial Graphing Node

d. Spatio-Temporal Prediction Node

e. Geospatial Association Rule Node

 Users can leverage these nodes to build powerful models factoring in space and time, as well as other cross sectional variables. To get an idea for how this type of modeling works, suppose a police force wants to predict the rates of different violent crimes so that they can better allocate units to different places at times when they are most needed. The types and frequencies of violent crimes change based on when and where they take place, so the space-time correlation needs to be modeled to get a better prediction. For example, in certain areas, violent crimes may have a higher likelihood of happening during daytime while in some other areas they would more likely happen after midnight. This same approach can apply to store location choice, traffic congestion prediction, epidemic spread prediction, and many other analysis types.

Conclusion

As you can see, there have been some substantial additions to SPSS Modeler in version 17. If you want to get started with this powerful predictive tool and begin acting on your future before it happens, Ironside can help. Get in touch and we’ll show you what this game changing technology can do for you.