There’s no doubt about it––the Internet of Things (IoT) is shaping up to be the mother of all technology trends in 2016. There are endless statistics on how many devices are or will be coming online; just within the home, everything from doorbells to refrigerators, to light bulbs has become internet-enabled.
As IoT devices infiltrate many product ecosystems, they’re becoming more a part of our lives. But while these devices gather lots of data, they’re not very intelligent or self-aware in their own right. That can be a problem. We need the right intelligence and security onboard, in the devices, in order for them to become aware when they might be manipulated or failing.
I call it the “Intelligence of Things,” and that’s when things will get really interesting for IoT.
The power of self-learning analytics
Self-learning streaming models based on continuous inputs are a mainstay of the financial world––in fighting payment card fraud, for example––and, increasingly, cybersecurity. As applied to IoT devices, self-learning models monitor their environment and gather data, and thus can make a determination if the behavior of a user or an environment they’re embedded in is normal or abnormal.
With self-learning models on board, IoT devices can warn of an unsafe situation or an impending failure. Stoked by streaming analytics, the device will infer that something about its environment or itself is failing, and issue an appropriate warning. Interestingly, IoT devices within the home can communicate with one another; in this way, the web of devices can build a larger consciousness of the house based on their collective inputs.
IoT spotlight: air conditioners
Here’s an example. In recent years, commercial air conditioning units have become targets for vandals, looking to strip the $80-$100 of copper contained therein. Some companies have taken to elaborate measures to protect their A/C units, ranging from lighting and security cameras to special alarms, to putting a GPS tracker on the unit.
Outfitting the air conditioner with an IoT sensor would allow it to collect a constant stream information on many operational functions. Should a component begin to malfunction, the IoT device could trigger an alarm, notifying the facilities manager that maintenance is required. In this way, should the device be intruded upon by vandals, the IoT sensor would detect abnormal activity and trigger an appropriate alarm.
Here’s another, even more, dramatic air conditioner example. In 2015, Formula 1 race car driven by Jenson Button and his wife were robbed in their villa in the south of France, reportedly after burglars pumped an anesthetic gas through the house’s air conditioning system. (Shockingly, this kind of attack does not seem to be uncommon in the region.)
Again, an IoT sensor could have detected tampering with the unit, or air pressure changes as the anesthetic gas were introduced into the system, triggering an alarm. The alarm from the air conditioning unit, coupled to inputs from the villa’s IoT home monitoring system, could have alerted the owners (or a private security monitoring company) of suspicious activity.
On a more pedestrian level, even the lowly IoT doorbell could benefit from self-learning. Earlier this year hackers figured out how to compromise the Ring IoT doorbell, to extract the home WiFi network’s password. Putting more intelligence on-board in the device could have triggered an alert that the doorbell was being tampered with. This information could’ve been crossed-checked with the motion-triggered alerts that Ring already supplies.
Drilling down on self-learning
Moving from vanilla IoT to the “Intelligence of Things” requires a change in the mindset of these devices. Instead of just collecting data for apps, comparison to rules thresholds, or binary commands, they need to self-monitor their state in the environment. They must measure their own “self” which, for an IoT device, requires streaming behavioral analytics.
“Behavioral analytics” has mixed meanings today. To most technical people the term typically refers to heuristics such as, “If X, Y, Z happen in that order, then action A will be taken.” These event chains, over time, have their place, but device self-awareness requires a unique understanding of individual environments; no two IoT devices are placed in exactly the same environment. In this sense, they are as unique as their owners’ and users’ behaviors.
Real-time behavioral awareness is a fixture in certain domains such as payment card fraud detection. Here, a small (1,000-2,000 byte) entity profile of recursively updated feature detectors for fraud is maintained. These entity profiles have a small memory footprint, which allows two important things:
- Real-time updates in milliseconds
- Highly predictive analytic variables, monitoring normalcy vs. abnormality, can be applied in the stream of data.
This same approach is well suited for IoT devices for which:
- It’s not feasible to store all sensor data onboard
- Having an onboard entity profile that is updated with each and every sensor read is definitely feasible, and allows the device to perform sophisticated self-inspection.
Determining outliers and appropriate actions
Each feature in this entity profile has a normal range of variation and can get very detailed, specifically, understanding which sensor data is interesting or relatively uninteresting. “Interesting” data would occur when a house’s inhabitants are away at school or work––or when they are in the home, depending on the IoT device and its application.
Distributions of these “interesting” and “uninteresting” features, computed in real-time, allow for a determination of which ones would be in an outlier state, and how extreme. The outlier features then can be combined to produce a score that can be operationalized as to actions the device has authority to perform.
Collectively, these techniques are well proven and have been utilized on smart devices for more than a decade to monitor components, detect utility network failures and signal infrastructure changes at a national level.
Achieving the “Intelligence of Things”
By leveraging the predictive power of self-learning analytics, IoT can facilitate exciting positive outcomes. “The Jetsons,” the 1960s cartoon TV show, has also been mentioned as a benchmark for the connected home. With the “Intelligence of Things” extending data-collecting IoT devices into an analytic fabric, we’re closer to a Jetsons future than ever.
The author is Chief Analytics Officer of FICO, a leading analytics software company, helping businesses to make better decisions. Follow him on Twitter @ScottZoldi