A vital but rare skill among technology management teams is the ability to prepare and apply tools that help an organization adapt to the present and future and not just respond to the recent past. BA (business analytics) and BI tools enhance an organization's ability to survive by facilitating a full range of defensive strategies, such as reducing risk or lowering inventory, and aggressive moves, such as finding new customers or higher-margin markets.
With that in mind, SAS Institute Inc. recently updated its data-mining dynamo, Enterprise Miner 5.1 for SAS9. The improved model-design options make a strong argument for bringing data mining into the enterprise.
The right questions
SAS' sophisticated view of applying BA to data mining and related tasks shows in Enterprise Miner 5.1. The product sits on the distributed client/server architecture of SAS' Enterprise Miner Server and can access the company's add-ons to SAS9. The new version uses multithreaded execution to distribute its work over multiple processors and to perform simultaneous calculations on multiple models in the client.
Although Enterprise Miner is a data-mining application, it includes both standard BI tools and an expanding set of BA tools. Enterprise Miner aids analysts in modeling business processes from existing data and predicts what might happen based on particular questions. The solution's toolset provides analysts with the ability to connect to data sources, to design or select input, and to massage or prep data for analysis -- all of which set this product apart.
The software supports the data-mining cycle in a thorough and, for the most part, elegant style. It starts with problem analysis, asking the right questions and knowing what kinds of answers can be acted on. The process continues with a work cycle SAS calls SEMMA (sample, explore, modify, model, assess), in which an analyst or statistician works directly with Enterprise Miner to deliver tested models that can be used as-is on the spot, be saved for reuse, or be exported for integration with other programs.
The Enterprise Miner client contains a set of tools for attaching to data sources. Through the client, analysts can tap external tools available in SAS9 or other existing mechanisms in order to prepare and clean data from data warehouses, data marts, or other significant databases.
SAS9 comes equipped with many native intrinsics and low-level programming methods for accessing and refining data. Also available are add-on products such as SAS/Access and ETL (extract, transform, and load) Studio for organizing data drawn from diverse sources before it is analyzed in Enterprise Miner.
Performing data analysis with Enterprise Miner is a straightforward process. After attaching to data sources, the analyst sets up a sampling routine, selects variables to target, and shapes data import by deciding how to deal with exceptions such as records that are missing values in specific fields.
The analyst can also refine the data types that the import intelligence guessed at and assigned at the time of attachment. The program offers tools for charting basic details about the universe of data. Those tools help the analyst select sampling approaches by rebalancing the incoming data against the results of specific fields.
I liked the way Enterprise Miner's intelligence extended through the workflow and how smoothly the client supported changes -- any analyst with beginning or intermediate statistics skills will find this process simple.
After a sample set of working data has been obtained, the analyst can explore the set graphically with simple, informative, but not spectacular charts. Although adequate, the charts won't add a lot of value for those analysts who are comfortable with summary numbers -- that is, most sophisticated users. The new version can, however, export diagram results as XML files for distribution to other analysts who may be collaborating on the solutions.
The next phase is to build a set of tasks to design a model that answers questions by selecting component "nodes." Nodes are dragged and dropped onto a diagram workspace, where the analyst connects them in a processing sequence.
Each node contains a package of functions for manipulating or analyzing data or for displaying results. The interface presents a chain of nodes that represents the sequence an analyst works through from input to output. It's an elegant and effective method that will increase both productivity and the ability to refine models.
When results have been obtained, the analyst can examine them graphically or as tables (or both) in order to gain insight into the processes. Those insights fuel a subsequent tuning of the processes or perhaps trigger new questions and modifications to the design that will enhance the model's ability to deliver better-tuned results iteratively.
Enterprise Miner 5.1's new node choices include an AutoNeural node that speeds the process of defining a neural network. Neural networks are useful but usually require a fair amount of work and a little bit of luck to tune. The new node accelerates the refinement of the algorithm, thereby eliminating much of the problem.
After an effective model has been created, the analyst can apply the model to incoming data by exporting it to SAS9 or as C or PMML (Predictive Model Markup Language). Enterprise Miner 5.1 now includes an experimental Java API, which supports embedding of the product's processes into custom programs so owners can extend the client utility.
Data mining is the one sure way to progress regardless of the economy's direction, but which vendor's solution works best remains an open-ended question. SAS' offering has certainly gotten stronger with this version of Enterprise Miner.
Its new program architecture enhances performance by speeding iterative work and expanding the number of questions and size of data sets that can be analyzed. Enterprise Miner 5.1 is a dynamo that should be an automatic upgrade for SAS shops. -- InfoWorld (US)
Join the CIO New Zealand group on LinkedIn. The group is open to CIOs, IT Directors, COOs, CTOs and senior IT managers.