Getting Started with Predictive Applications – Buy Versus Build
Date: December 21, 2015
Research Note Number: 2015-49
Author: Jim Lundy
Issue: What are the key trends in predictive analytics?
Summary: Predictive Applications that leverage Machine Learning and Cognitive Computing are here to stay. Enterprises can purchase off the shelf Predictive Applications or build their own.
Computers and humans will work closer together than ever before thanks to new predictive capabilities. Figuring out how to leverage predictive is crucial in the coming era of Smart Applications (See Note 1). This research note provides an overview of how to get started with predictive applications, using Machine Learning.
Predictive Applications – Planning is Critical
To make software smarter, you have to teach it about data. Machine learning algorithms make next generation software applications smarter. One of the things an enterprise needs to decide is what applications it wants to make predictive.
Advanced analytics used to be difficult and time consuming. In the past, legacy analytics programs required data scientists to sort data, write models, test models, and to visualize and interpret the results. It used to take months to go through a single data set. Today, the advancement of cloud-to-cloud technology and the rise of machine learning algorithms are dramatically changing how software reacts to changing conditions in real-time and how it makes appropriate suggestions to humans. Making software predictive requires teaching the software to look at historical data, to ensure that it understands the past.
Furthermore, cloud-to-cloud technology makes data access far easier. Users can feed data from any cloud they use into the cloud that stores their analytics. There is no need to change data types or locations, since platform providers have been working on making their systems more connectable. Additionally, the rise of machine learning has also made analytics easier. After a short training period, where users run historic data into a new model so it learns to identify trends, the analytics are then able to run themselves.
When planning for predictive analytics, an important thing to account for is the owner of the data. Data is the most valuable commodity in analytics, and retaining ownership of it is important. You should be aware of the data’s origin and what it might be used for in the future. For example, if you wrote a predictive program that looked for sources of fraud, trained the model with historic data, and then sold it, do you still own the data? Or does the person who purchased the program have a claim to that data because it was a necessary part of the model they purchased? In the era of predictive, data ownership will be everything; any enterprise that uses predictive should keep a close eye on the location and ownership of their data.