Machine learning is all about automating the development process for analytical models. One way to extend the use of machine learning is to broaden your library of machine learning algorithms. Another way is to scale your machine learning process by reducing the time required to process machine learning algorithms on large and complex data.
Every day, I hear organizations asking:
- How do we get people with the skills to apply machine learning successfully to solve business problems?
- How do we integrate analytics into our business processes, helping us get data-driven answers from these models to decision makers at the right time?
- How do we create an analytics culture where all decisions are based on the analysis of data?
- How can our analytical talent collaborate more efficiently?
Each of these questions can be addressed by automating machine learning and increasing its use.
Automated machine learning at scale
What you need is an automated predictive modeling process for testing different machine learning scenarios quickly, finding champion models easily, and integrating these models into production seamlessly.
What if your users could create and run model tournaments with just a few clicks? Finding the best model for your business problem is easy when you have neural networks, decision trees, random forests and other machine learning algorithms competing against each other.
Of course, sometimes you also need to open the lid and fine tune parameters of the modeling process template. You need the flexibility to turn on the autopilot for machine learning where appropriate and take control of the engine when required.
Even more machine learning
For many years, data miners have used a growing library of machine learning to maximum control in analytics projects. Now, machine learning capabilities can be added to high-performance Bayesian networks too. Data scientists prefer Bayesian networks because they are relatively easy to compute and the results are easy to interpret.
With unsupervised modeling, or clustering, the key question is: How many clusters are inherent in my data population? Data miners can now apply a new method in high-performance clustering to automate the number of clusters selected in large data sets, using the Aligned Box Criterion (ABC).
These capabilities are now available in the latest version of SAS® Enterprise Miner™.
Another area of interest is the automated analysis of unstructured data to observe trends, spot new topics, issue alerts about potential problems and flag new business indicators.
New machine learning techniques can identify patterns in unstructured data and automatically create Boolean rules to describe the patterns. Users of SAS® Text Miner™ and SAS® Contextual Analysis can refine and overwrite the machine-generated rules, enabling semi-supervised, active learning and dynamic interaction with the algorithm. The software automatically learns categories and topics from the collection. Users can then guide the system to an improved solution – enabling interactive model building.
For maximum performance on a single server, the Boolean rule extraction now also supports multithreaded processing to take advantage of multiple core processors. And – these high-performance, automated learning algorithms now support 13 different languages: Chinese, Dutch, English, Finnish, French, German, Italian, Japanese, Korean, Portuguese, Russian, Spanish, and Turkish.
How are you using machine learning – or how would you like to use it in the future? We would love to hear from you about your thoughts on machine learning.
Learn more about SAS Analytics.