What is Machine Learning?
We begin with two assumptions. First, that you’ve heard of machine learning before. And second, that you may not know exactly what it is or how it works.
Machine learning has been a buzzword of astronomical proportion over the past few years. It promises a revolution in the way we analyze and consume data. To yield trends and outcomes invisible to a human mind. But for all the hype, few truly understand what it is, how it works or the full scope of impact it promises to have on sectors like the events industry.
So…just what is machine learning?
“Machine Learning is the science of getting computers to learn and act like humans do, and improve their learning over time in autonomous fashion, by feeding them data and information in the form of observations and real-world interactions.” – Dan Fagella, ‘What is Machine Learning’ (2012)
To understand what machine learning is, it’s best to start with the results that can be produced in the end and then examine how it’s achieved. In its most pure form, machine learning allows for two types of outcomes when analyzing data:
– Use of data to form predictions.
– Use of data to discover and show similarities and trends.
In other words, machine learning makes predictions based on past indicators and identifies new ways to structure data based on similarities within that data.
For example, in the event space machine learning can use specific attendee behaviors to indicate a higher chance of returning to the event the following year. Or it can take entire event datasets and segment attendees into more specific groups based on previously unrecognized attributes.
What’s Under The Hood?
To achieve prediction results like the likelihood of an attendee returning to an event, a machine learning model must first be created. Essentially, a model is combination of inputs, desired outputs, and the algorithms to help define correlation.
Structured data that indicates a value that a machine learning model will analyze. This could be anything from a yes/no value (did the participant attend the education session they registered for) to a number (how did the attendee rate a session from 1-10).
Classifications of what a specific set of data indicates. As a very simple example, take the known input samples from above. Say that you know attendees who attend 80% of the sessions they’ve registered for and have an average session rating of 7/10 or more are classified as ‘likely to attend’ next year. If those numbers are between 60-79% and 5-6/10 respectively, the attendee is classified as ‘less likely to attend’. And if those numbers are lower, then the classification is ‘will not attend’.
Iterations & Time:
To improve models, iterations of a dataset over time are required to craft more accurate results. Attendee information, for example, is collected year over year and consistently feeds the model.
Supervised Learning vs Unsupervised Learning
Machine Learning processes data in two ways: supervised learning and unsupervised learning. These processes achieve different results based on analysis of the data.
Supervised learning forms similarities in data into pre-determined predictions. Take the example used to explain know outputs above. This is a supervised learning scenario in which the inputs- session sentiment and attendance- are used to help classify attendees into a list of pre-existing categories- likeliness of reattendance. Ultimately, the goal of supervised learning is to make predictions based on structured data.
On the other hand, unsupervised learning analyzes data to find and group similarities into clusters but does not make predictions. The human mind does not effectively process data sets that are ideal for machine learning. This results in missed opportunities to better structure and defines known outputs. Unsupervised learning is exploratory by nature and extremely effective in helping identify new clusters of data that only a computer could discover but help in shaping known outputs.
Take once again the reattendance prediction example from above. The known outputs- ‘likely’, ‘less likely’ and ‘will not’- were based on several basic attendee data points. The inputs of session attendance and session rating offer obvious correlation to the human mind. But what happens when the known inputs deviate from the data required to form a known output. Say an attendee joins each of the sessions they’ve registered for, but rates each as a 3/10. Where does that fit into the model? Moreover, what if subtler, less connected data was indeed just as important in helping form predictions?
Data points like how early an attendee registers for the event, their personal attributes, on-site actions like exhibits visited and many other areas in the data could be just as impactful, if not more important, in determining known outputs.
These clusters would often not be discoverable- and thus not used in modeling- if not for unsupervised machine learning. When discovered, these clusters are used in future iterations of the model to produce better-known outputs.
Potential Impact on the Events Industry
Machine Learning’s potential impact on the events industry isn’t hard to imagine. Predicting the rate of return for attendees is just one of countless scenarios in which machine learning can take large datasets and produce more accurate and actionable insight. It can answer straightforward questions like how many registered attendees will actually show up to the event. Or power sophisticated models that show where to place marketing budget.
Moving forward, event teams should expect to see an emergence of machine learning in the software they use to help optimize and improve the way they analyze and plan events. And for those who can use it effectively, machine learning will transform from a buzzword into an effective way to level up their events and conference like never before.