Internet of Things

Machine learning will bring out the best in IoT…

Recently, IoT Agenda shared three perspectives on how to navigate the challenges of IoT. I was honored to contribute my thoughts on managing massive amounts of IoT data. While my portion is reprinted below, be sure to check out the original column for additional perspectives from industry experts.

One of the critical challenges of IoT is avoiding death by data. IoT generates tremendous volume and velocity of information, easily swamping most homegrown data collection systems. Additionally, organizations find it difficult to sift through the data to find actionable insights in a way that is both scalable and forward-looking.

The first hurdle is surviving the onslaught of incoming data. In recent years, this barrier has been dramatically reduced thanks to advancements in streaming and storage from major cloud providers.

Once the mechanics of ingesting and storing are mastered, the real challenge begins: How do organizations breathe life into the data?

A common knee-jerk next step is to develop dashboards, dashboards and more dashboards. There is nothing wrong with building out a strong suite of analytics, but caution needs to be heeded. IT pros can use modern visualization tools to slice and dice data at ease. They also make it easy to get sucked into a vortex of pie charts and bar graphs that look pretty but provide only incremental knowledge. Just because IoT gives all of the data, does not mean IT pros need to use all the data. Be thoughtful: What is the heartbeat of what the IoT project is trying to accomplish?

Extending the metaphor to a healthcare example, having sensors throughout a hospital gives healthcare providers the ability to track everything from temperature changes in rooms to how often the trash is taken out. But will creating elaborate dashboards on all those data points improve patient care? Likely not. But surfacing something like the frequency and length of patient and caregiver interactions — a known driver of improved outcomes — will. By culling the extraneous data and reports, users avoid being overwhelmed and, more importantly, surface the most valuable insights.

That said, IoT truly comes to life when paired with machine learning. Machine learning pushes IoT to transition from reactive to predictive analytics, forecasting future outcomes and cutting through the fog of data. Machine learning also thrives on scale, making it a natural pairing for voluminous IoT data.

Extending the hospital example, adding in machine learning transitions the conversation from, “How many patient and caregiver interactions were there last month?” to, “How will next month’s staffing levels affect patient care?”

However, machine learning is not inherently easy to execute. Building and deploying custom models for IoT data requires a strong understanding of both big data and data science. Thus, organizations often delay or outright skip machine learning. Recent advancements in this space are promising, but IoT will not reach its full potential until machine learning is more accessible.

Leave a Reply

Your email address will not be published. Required fields are marked *