Should I Supervise My Machine Learning?
There is a lot of buzz about artificial intelligence (AI) and machine learning (ML). Technically, machine learning has been around for many years. It’s now moving more into the mainstream world solving business problems in many areas. Let’s take a brief look at the use cases for three different types of machine learning.
The first groups to apply intelligence were the marketing organizations. We were all impressed with the recommendations we received years ago, but the early versions were rules-based engines that were based on predetermined outcomes. If a client buys razors, then suggest shaving cream. Not really effective but better than nothing.
Supervised Machine Learning
The next technology leap was combining advanced analytics with rules-based systems. Advanced analytics created more sophisticated recommendations based on past transactions and behavior of similar individuals. In essence, this is Supervised Machine Learning. Data scientists build models then embed the algorithms into applications to execute automatically. This approach works well when you know the outcome you are looking for and the data is relatively static.
Unsupervised Machine Learning
Unsupervised Machine Learning is the best approach for data that is unstructured such as voice, video, images, and text. The system analyzes the inputs and learns as it goes. This ability to learn has created a lot of excitement and provides solutions for problems that couldn’t be addressed previously.
This process works best for images that may be evolving and different from previous references. There are many applications in sales and services platforms that can leverage these capabilities. For example, Unsupervised Machine Learning can scan trouble tickets and determine the best group to address the problem or even suggest the probable fix itself.
Semi-supervised Machine Learning
Another variety of machine learning is semi-supervised. This doesn’t have the full learning capabilities of Unsupervised, but there are many use cases that could benefit from this approach. Semi-supervised learns from the data but it is instructed by the user of the desired outcome. Semi-supervised ML leverages similar algorithms as Supervised but enables the system to train the models as new data is ingested, improving speed to results.
This approach works well for business problems with dynamic data and changing scenarios to consider. Sales and marketing organizations have leveraged Semi-supervised machine learning for years for business challenges that require many tailored models.
One example of this could be online marketers that want to achieve personalization—the process of making the customer feel the offer or engagement was created specifically for them. Operations and engineering departments are now starting to adopt this approach to predict quality impact on products and projects that may be dynamic in nature.
Which Is Right for You?
The right approach depends on the problem to be addressed. There may be scenarios when you use all three. Supervised Machine Learning for the static, well-defined problem that requires the perfect model. Unsupervised to incorporate all of the images, video, and voice generated in today’s connected world. And semi-supervised to accommodate the changes and generate numerous models needed to compete in today’s environment.
Having said all that, the key is incorporating the machine learning results into applications that can affect the business process.
Benefits are not derived from machine learning results. Benefits are derived from the process changes individuals can make by implementing these results immediately.
Learn more about predictive analytics topics by reading the other blogs in our machine learning series.