This Game of Thrones algorithm predicts who will die next in Season 8

This Game of Thrones algorithm predicts who will die next in Season 8
Who is most likely to survive? (L-R): Jaime Lannister, Daenerys Targaryen or Bronn?HBO

  • A team at the Technical University of Munich developed an algorithm in 2016 that can predict who's going to die next on the Game of Thrones.
  • Not only can the algorithm predict who is most likely to survive and who is most likely to die, but also the likelihood of death depending on whether a character is female or male, a noble or peasant etc.
  • NO SPOILERS

One of the enticing aspects of Game of Thrones is that you never know who’s going to die next — and, it’s usually when a character is beginning to grow on you. And, Season 8 of Game of Thrones has a lot of strong contenders for the hit list.
TOP VIDEOS FOR YOU

So, to spare viewers the heartbreak, a team at the Technical University of Munich (TUM) came up with an algorithm that will answer the ever burgeoning question of, who’s next?

Advertisement

The epic scale of the worlds created by George R. R. Martin provides an almost endless resource of raw multi-dimensional data. It provided the perfect setting for our class.

Dr.Guy Yachadav, who led the class and conceived the project

First are foremost, the algorithm was able to determine that there are some basic principles in the Game of Thrones universe. For instance, if you’re a woman — then you’re more likely to die, with a 21% death rate compared to 11% for men.

This Game of Thrones algorithm predicts who will die next in Season 8
Predicted likelihood of death between male versus female Game of Thrones charactersgot.show

And, between nobles and peasants, peasants are definitely more likely to die than the royalty.

Advertisement

This Game of Thrones algorithm predicts who will die next in Season 8
Prediction of likelihood of death between nobles and peasant characters on Game of Thrones

There’s also a correlation between the house that a character belongs to and their likelihood of survival. Along with the many privileges of belonging to the House of Lannister, a Lannister character is also 45% more likely to survive than characters that belong to other house — like being a Baratheon, where they are 5% more likely to die than the average character of the series.

But, who’s next?

As of now, the queen of the dragons — Daenerys Targaryen — has the highest probability of survival with only a 0.9% chance of death. "The cleverest man alive" is also safe for the time being with only a 2.5% chance of dying.

Advertisement
So far, the algorithm has exhibited a 74% success rate of its predictions in pegging which characters are relatively safe and which ones are in grave danger — like the remaining children of the House of Stark. Sansa and Bran, both, have a more than a 50% chance of dying next in the Fire and Ice series.

This Game of Thrones algorithm predicts who will die next in Season 8
Who is most likely to die next on the Game of Thrones?

But fans of Bronn, played by Jerome Flynn, have more to worry about since the likelihood of his death is at a whopping 93.5%.

Gregor Clegane, or better known as ‘the mountain’, is also among the list of characters that may die this season with a 80.3% chance of not surviving what comes next.
Advertisement

Predicting which GoT character will die next

Only the fewest of characters die of old age, so the TUM algorithm uses machine learning (ML) that allows their computers to make predictions using sufficiently large number of examples from the past to compile statistics.

These statistics, in turn, predict what’s going to happen next. The Wiki of Fire and Fire and the Game of Thrones Wiki give the algorithm the data it needs, which is then plugged into a Bayesean survival analysis that works on an MCMC simulation with a pymc3 package.

Advertisement
And then, the team trained a neural network to predict the probability of outcomes using the Python’s Keras framework. It’s basically one the easiest neural network architecture that uses the Feed Forward technique — which is, essentially, processing the data, after its fed to the system as a vector, through ‘hidden layers’.

Through this entire process, the team is able to prediction transitions and changes in the percentage likelihood of death over time.
{{}}