As businesses flock to the Internet of Things (IoT), connecting everything in sight and adding new functionality to old hardware, they're all really after only one thing: Data.
Data is the new oil, and an enterprise IoT deployment is an easy way to get a lot of it from many different sources. But in the end, it's what a business does with its data that really matters. By putting that data to work, organizations can improve efficiency, boost the bottom line, and drive innovation.
That's where machine learning comes in. It's still a fairly nascent technology, but some companies are using machine learning to boost the value of their IoT initiatives.
For starters, machine learning algorithms can make IoT data better suited for processing and analysis. Businesses can use trained algorithms to help in the organization and tagging of data. Using machine learning, a company can determine data provenance and classification, as well as if it meets certain requirements for compliance. This could be especially helpful for IoT deployments in the highly regulated medical and financial services fields.
Machine learning can also be used for the analysis itself. The technology can "provide predictions, recommendations or potentially prescriptive actions" for enterprise IoT deployments, according to Dave Schubmehl, research director for cognitive and AI systems at IDC.
SEE: IT leader's guide to deep learning (Tech Pro Research)
The core use case for this type of algorithm is predictive maintenance. This is done when sensors on complex machines send data back that's "used to predict when various sub-systems might fail and recommend when that machine should get preventative maintenance to keep failures from occurring," Schubmehl said. By using data to address maintenance issues before failure occurs, businesses can save time and money.
Predictive maintenance use cases represent about two-thirds of the IoT deployments that 451 Research sees, according to Christian Renaud, the firm's research director of IoT.
What typically happens, Renaud said, is that "there's a lot of real-time data that you're monitoring, but you don't really start capturing and analyzing until there's an exception case." One example would be a hospital with high-value refrigerators for organ transplants that need to stay at a constant temperature. No-one really cares that they stay at a constant temperature until they don't anymore, and machine learning can (hopefully) keep them from failing.
Resource management is another way machine learning can be used with IoT initiatives. According to Schubmehl, companies like John Deere use "sensors on tractors and farm equipment to monitor the state of the soil, plants, insects, moisture, etcetera in order to build predictive models to gauge exactly how much fertilizer, water and pesticides should be applied in order to maximize crop yield."
In the 2017 Gartner report AI on the Edge: Fusing Artificial Intelligence and IoT Will Catalyze New Digital Value Creation, an example was given of how Google uses IoT and machine learning to optimize the resources in its data centers. According to the report, sensors monitor temperatures, power, pump speeds, setpoints, and more. Using that data, and a specific algorithm, Google reduced its cooling bill by 40 percent and got 3.5x computing power from the same energy consumption.
Data collected from radio frequency identification (RFID) tags can also be used with machine learning to create business value. Schubmehl gave the example of RFID used in the shipping industry to optimize routes and logistics for supply chain. Renaud said this is seen a lot in trucking, where machine learning is used to determine which route impacts the engine the least and helps maintain the best fuel economy.
Currently, machine learning implementations in IoT are more prevalent in mature verticals like manufacturing and transportation that have been working with these kinds of technologies for some time, Renaud said. Most companies, however, are "still in a period of experimentation," where they don't know what the significant variables are yet, Renaud said.
As machine learning further comes into its own in the enterprise and in conjunction with IoT, other new use cases will present themselves. One of these use cases will be machine learning used to understand contextual customer data.
"You're going to get a lot of user intent from things like omnichannel marketing for retail -- being able to tie you as a consumer and your online behavior to what you do in-store," Renaud said.
The Gartner report also mentioned contextual data in retail, specifically using in-store video cameras and machine learning to create intelligent video analytics.
Other integrations of machine learning and IoT include ingestible autonomous surgical robots, using manufacturing data and algorithms to autonomously trigger other specific actions in the manufacturing process, and using connected vehicle data and machine learning to create customized insurance offerings, Schubmehl said.
While it's impossible to predict all the ways that machine learning will impact IoT, it's pretty much a foregone conclusion that machine learning will be the linchpin that drives real enterprise value in IoT.
- Special report: Harnessing IoT in the enterprise (free PDF) (TechRepublic)
- Machine learning and the Internet of Things (ZDNet)
- Machine learning: A cheat sheet (TechRepublic)
- Analytics in 2018: AI, IoT and multi-cloud, or bust (ZDNet)
- 5 tips to overcome machine learning adoption barriers in the enterprise (TechRepublic)