Skip to Content

Machine Learning Algorithm Predicts Which New Faces Will Make It as Fashion Models

A machine-learning algorithm picks out the fashion models most likely to succeed.

The job of choosing fashion models to represent a brand has never been easy. When searching through any database of models, a casting director is faced with a choice of thousands for every magazine cover, social event, or runway appearance. Which to choose has become more complex in recent years because social media now plays an important part in any model’s career.

And yet some faces quickly become more popular than others. That implies that when faced with the same information—usually things like body size and shape, model agency, previous experience as well as an image of the model—casting directors generally seem to make the same choice.

That raises the interesting question: given the same information, could a suitably trained machine make the same choice and even predict which models are more likely to appear on next season’s runways?

Today, we get an answer thanks to the work of Jaehyuk Park and pals at Indiana University in Bloomington. These guys have used a machine learning algorithm to spot the factors that correlate with future modeling success, as measured by number of runway appearances. And they say their approach becomes even more accurate when it takes into account social media popularity as well.

The team began by downloading the data associated with 431 female models who were listed as “new faces” on the Fashion Model Directory, an important industry listing website. This site gives various details about each model, such as name, age weight, height, hip, waist, dress size, and so on. Unsurprisingly, these data show that on average these models are significantly taller, thinner, and lighter than the rest of the population.

The Fashion Model Directory also lists each model’s agency—which the team categorizes as either a top agency or not—and her experience to date, such as magazine covers and number of runway appearances.

Being new faces, these models all had a similar, limited level of experience with, on average, just 3.25 runway appearances each during the September 2014 fashion weeks in New York, London, Paris, and Milan. That figure is deceptive given that the majority of them did not perform a single runway during those weeks and only 24 percent performed one or more.

The team also collected data about the presence of each model on Instagram, probably the most influential social network in the fashion world. They found that almost 60 percent of the models had Instagram accounts with those at top agencies more likely to be represented on Instagram.

Park and co went on to collect all the social posts for each account in the three months leading up to the September 2105 fashion weeks, including metadata such as the number of likes and comments. They even calculated the sentiment of the comments to determine to what extent they were positive or negative.

Having gathered all this data, they used various machine learning algorithms to look for correlations between those that had one or more runway appearances and those that had none.

The results make for interesting reading. Various factors positively correlate with runway popularity—for example, tall models are more popular and each additional centimeter of height more than doubles their chances of walking a runway. Being on the books of a top agency is an even more important factor which increases the chances of appearing on a runway by a factor of ten.

Rather predictably, factors such as larger dress, hips, and shoe size all negatively correlate with success while waist size is not correlated either way.

Social media turns out to be important too but not always in obvious ways. More comments on an Instagram account correlates with higher chances of walking a runway. But strangely, having more “likes” reduces the chances by around 10 percent.

Now here’s the significant part. Having discovered these correlations, Park and co use them to attempt to predict success at a future set of fashion weeks, specifically those in February and March 2015.

Once again they trawled the Fashion Model Directory for new faces prior to these shows and downloaded the data associated with 15 models along with their Instagram data. (Why there were so many fewer new faces this time isn’t clear.)

Finally, they used their machine learning algorithms to predict which of these models would do one or more runway appearance and which would do none.

The best algorithm correctly identified six of the eight models who would go on to become popular on the runway.  (Indeed the algorithm correctly predicted that all the models in the images above would become more successful, to answer the question in the subtitle.) “Our framework successfully predicts most of the new popular models who appeared in 2015,” say Park and co.

The team also analyzed which factors were most important in these predictions and found that social media plays a key role. “We find that a strong social media presence may be more important than being under contract with a top agency, or than the aesthetic standards sought after by the industry,” they say.

The study has a number of weaknesses, however. The most serious is that the team demonstrates this predictive power for only 15 models, something they hope to address with bigger data set in future.

There is also a question over the team’s measure of success—a runway appearance. Any model will tell you that runway appearances are not created equal with those associated with the biggest fashion houses, such as Chanel and Hermes, being far more important and valuable. Again this is something the team hopes to address in future work.

Another limitation is that the results apply online to female models due to the lack of available data on male models.

Nevertheless, Park and co provide some interesting insight into an industry that is largely opaque. It also contributes to a broader field of endeavor in the form of the emerging field of the “science of success” which studies the forces and mechanisms at work when individual become successful or fail.

The fashion world is notoriously ruthless. “We note how fashion modeling exhibits a strong winner takes-all component,” say Park and co. “In an industry that seems to be governed by such a survival of the fittest mechanism, the difference between performing a show in a premier venue or not becomes crucial.”

In these kinds of scenarios, even the smallest advantages become hugely amplified. And that has important implications for any budding supermodel—a strong Instagram account could be the difference between future success and failure.

Ref: arxiv.org/abs/1508.04185 : Style in the Age of Instagram  

Keep Reading

Most Popular

Large language models can do jaw-dropping things. But nobody knows exactly why.

And that's a problem. Figuring it out is one of the biggest scientific puzzles of our time and a crucial step towards controlling more powerful future models.

OpenAI teases an amazing new generative video model called Sora

The firm is sharing Sora with a small group of safety testers but the rest of us will have to wait to learn more.

Google’s Gemini is now in everything. Here’s how you can try it out.

Gmail, Docs, and more will now come with Gemini baked in. But Europeans will have to wait before they can download the app.

This baby with a head camera helped teach an AI how kids learn language

A neural network trained on the experiences of a single young child managed to learn one of the core components of language: how to match words to the objects they represent.

Stay connected

Illustration by Rose Wong

Get the latest updates from
MIT Technology Review

Discover special offers, top stories, upcoming events, and more.

Thank you for submitting your email!

Explore more newsletters

It looks like something went wrong.

We’re having trouble saving your preferences. Try refreshing this page and updating them one more time. If you continue to get this message, reach out to us at customer-service@technologyreview.com with a list of newsletters you’d like to receive.