Skip to content

What Sherlock Holmes has to do with data science

What Sherlock Holmes has to do with data science

It wasn't an action that caught Sherlock Holmes’s attention, but the absence of one. It was the fact that the dog didn't bark when the perpetrator entered. With that essential detail, the observant detective reported to the inspector that the crime must have been committed by an insider.

Sir Conan Doyle’s cunning character was a mastermind of context clues – those drawing a line to a certain context or those situationally out of place. In many ways, artificial intelligence (AI) seeks to connect dots in reality like the world’s most famous sleuth did in fiction.

Have any of these things happened to you in the lab?

  • A spike in freezer temperature
  • A midnight anomaly
  • A drop in tank pressure

Well, AI can deduce why they happened.

  • Someone simply left the door open
  • The freezer was likely undergoing a defrost cycle
  • The CO2 tank is the true perpetrator – and it’s empty

Context is everything. It isn’t enough to know that at midnight the temperature has dropped or the humidity has increased. Not if one cares to maintain a healthy sleep schedule, that is. Without context, a 3 a.m. temperature spike could be cause for alarm. But given smart context, an unnecessary fire drill can be avoided.

Replicating human capacity is just the beginning

AI’s lowest-hanging fruit is replicating human thinking. But its most disruptive action is drawing connections and uncovering insights that human brain power cannot. That discovery process is called unsupervised learning – learning beyond human limits and therefore void of human coaching.

In case you’re new to AI, here are the basics of what you need to know – and if this is rudimentary to you, please feel free to skip back to the juicy part:

  1. Weak AI (aka Narrow AI) is trained to perform specific tasks. Despite its unfortunate name, this type of AI is extremely powerful. It powers Apple’s Siri, Amazon’s Alexa, and even self-driving cars!
  2. While deep learning and machine learning tend to be used interchangeably, in reality, deep learning is a subtype of machine learning – and both are types of AI
  3. Deep learning can used supervised (labeled datasets to classify data or predict outcomes accurately) or unsupervised (unlabeled datasets to discover hidden patterns in data without the need for human intervention)

Elemental Machines’ machine learning employs both supervised and unsupervised learning to unearth insights like door open events that could otherwise be misinterpreted as freezer malfunction.

Due to the wide variation in how different equipment makes and models behave, making such deductions is only possible through finely tuned machine learning and copious amounts of diverse data.

The making of a model

To draw a single deduction like a door open event, Elemental Machines’ model takes into account both the expected behavior of an asset – like a freezer or incubator – and the historical behavior of the given asset. In the case of a refrigerator or freezer, compressor cycles, defrost cycles, heating elements, and fans can all impact the temperature profile of a unit. Each of these normal temperature fluctuations needs to be distinguished from the door-open event. By tracking the behavior of a freezer over a period of time, the unsupervised models can intelligently adjust their parameters to the specific behavior of any given freezer.

“Smart Context” represents the next level of inference. Using freezers as an example, there are a multitude of events that cause the temperature to fluctuate. Compressor cycles, door open events, power failures, and defrost cycles can all affect the temperature.


The Elemental Machines data science team uses the aforementioned unsupervised learning techniques to differentiate temperature fluctuation patterns and accurately label them. Then, leveraging the vast amount of time-series data, the results of the model over time builds confidence in the accuracy of the anomaly detection and is used to provide users with Smart Context Alerts. These alerts inform the user of abnormal fluctuations and the likely reason for the variations. This context allows operations professionals to triage alerts, save time, reduce stress, and ultimately save money by focusing on real problems.

Sherlock Holmes would be proud.

Learn more about Smart Context Alerts


This is part two of a three-part series on data science from Elemental Machines. In subsequent posts, you’ll learn how Elemental Machines’ data science-focused platform has helped others optimize operations and accelerate science.

Search our content