Who’s going to die in Game of Thrones season 8?

maarten-van-den-heuvel-73124.jpg

Two weeks ago we came to you with this post bemoaning the fact that there was no way we could predict who was going to die in the last season of Game of Thrones. We arrived at this conclusion using data we had on the TV show characters. But we enjoy a challenge so instead we turned to the data we had from the books. And guess what? We had a breakthrough!!!

 
Game of thrones predictions.png
 

You can skip straight to the conclusion for the spoiler if you like or, better, you can bear with us to see how we made our discovery. Let’s start with the methodology. We started by training a model - a weighted combination of information - on past data, in order to apply it to current data. For the purposes of this experiment we had the descriptions of many characters from the Game Of Thrones books in our data set, and we knew whether they had died, and in this case, in which book they had died. Our plan was to build a model to detect patterns among all the available data (books 1 to 5) that might announce an imminent death, and then apply it to the characters remaining to predict who is likely to die in the next book. But data mining projects require an important stage of testing before we can apply them, in order to ensure the quality of the predictions. This is why we built a first model on all the characters from the first four books (A Game of Thrones, A Clash of Kings, A Storm of Swords and A Feast for Crows), and then applied it to all the characters still alive at the beginning of book 5 (A Dance of Dragons). As we knew which characters died in the 5th book, we were able to evaluate if the predictions for the 5th book, based on the model built on the first 4 books, were significant.

 

The answer is YES! It is not the best model ever but certainly better than randomly choosing who will be the next character to be sentenced to death.

 
model predicsis gain chart.png
 

This was a good sign that the work was worth pursuing. The next step was to build a model on all the characters from the 1st to the 5th book inclusive. This would then be used on all characters remaining at the end of the 5th book, in order to rank them according to their likelihood to die.

⚠️ Here follows the big spoiler!!!

 
2GoT.png
 

The higher the character’s score, the more at risk he is of dying. For the analysis to be relevant, we should remember the timeline between the novels and the TV show. Counter-intuitively, the TV show developed ahead of the published novels. The last book is the 5th, whereas what happened in the last TV show season (7th) has not yet necessarily been written by George R.R. Martin.

See http://joeltronics.github.io/got-book-show/bookshow.html for further details. This means we can assume with reasonable confidence (although as far as Game of Thrones is concerned, ‘confidence’ is maybe a little presumptuous) that deaths happening in the TV show will occur in the final books. This is where we can partially check our ranked list: what were the positions, in our ranked list, of the characters who met their deaths in the TV show but not yet in the books?

  • Ramsay and Roose Bolton: 2nd highest score
  • Petyr Bealish (“Littlefinger”): 5th highest score. He’s not really high on the list, which confirms that his death was unexpected, based on other characters’ previous deaths.

Our joker is Tormund Giantsbane. He was left facing a burning collapsing wall at the end of the latest episode. Will he survive these events? As the first on our list of victims, we would wager he won’t.

To make things clear, we can point to two groups of endangered characters, and 1 group of safe ones.

 
Predictive Game of thrones
game of thrones predictions
 
 
game of thrones predictive analytics
 

With special attention to Jaime Lannister, who has the smallest chance of dying (0.06). Does that mean he’ll be the one to gain the Iron Throne…?

 

Alea jacta est, we’ve thrown our dice. What about you? What would be your bet? Send us your own predictions at contact , and let’s get back together in 2019, to confirm or refute your and our predictions. 👋


Intrigued? To get a sense of what we do at PredicSis.ai, please visit our Free Trial page