Predicting the Winner of Survivor Ghost Island

P

It’s time for another season of Survivor and that means I’ll be back each week to discuss what machine learning has to stay about the potential winner of the season.

If this is your first time here, you can read all about the details of the project here.

The quick and dirty version is that we use volunteers to collect data about player behaviors on an episode by episode basis, then using the data from existing seasons we train a machine learning classifier to calculate the probability that a given player after each episode is a likely winner or not.

Last season, this machine learning approach ranked the eventual winner in the top 2 the entire season and had Ben locked in as the winner starting with episode 7. That was a fantastic start, but who knows, maybe we got lucky.

Season 36 will be our second attempt to test this model.

There’s been a few updates since last season. Due to the buzz that we were able to create between podcasts and this blog, we had a huge number of data contributions since last year. We’ve added 27 season codings since last fall and now have coverage for every season except for two.

For the current season, we currently have 14 people registered to help code the player’s behaviors after each episode. I will use these submissions to update predictions each week and also try to provide some insights into what our model says about play style, outlying behaviors, as well as probability rankings.

So, without further adieu, let’s jump into it.

Winner Predictions Episode by Episode

For these predictions, I trained a Naive Bayes classifier based on 33 previous seasons of data. Although it might be useful to eventually consider things like idol finds, immunity wins, sex, age, and lots of other factors, for the training, I relied completely on the behavior data that volunteers sent me.

Week 1:

Below are the player’s ranked by highest probability to win after one episode.

  1. Domenick Abbate
  2. Stephanie Johnson
  3. Brendan Shapiro
  4. Michael Yerger
  5. Wendell Holland
  6. Chris Noble
  7. Morgan Ricke
  8. Kellyn Bechtold
  9. Sebastian Noel
  10. Desiree Afuye
  11. Laurel Johnson
  12. Bradley Kleihege
  13. Chelsea Townsend
  14. Libby Vincek
  15. Angela Perkins
  16. Donathan Hurley
  17. James Lim
  18. Jenna Bowman

Quick Thoughts:

These rankings are based on the complete 2 hour episode premier. Anything outside of the top 10, take with a grain of salt since many players were not shown in this first episode at all.

The top 5 looks reasonable. There’s certainly a chance that Domenick ends up burning out, but he found an idol in the first episode and he and Wendell seem to be working together. Stephanie had an outstanding first episode, seemingly controlling both the first two boots. Brendan and Michael appear to be working together as well and are observant and strategic.

Who’s playing like who?

For the first episode, I wanted to take a deeper look at the top 5 and see what prior winners they are playing like. I did this by taking each player from season 36 and calculating the dot product between their behavioral vector and all the winners. The combination with the highest dot product is the winner they are most similar to.

The results are below:

  1. Domenick Abbate most similar to Tony Vlachos
  2. Stephanie Johnson most similar to Earl Cole
  3. Brendan Shapiro most similar to Rob Mariano
  4. Michael Yerger most similar to J.T. Thomas
  5. Wendell Holland most similar to Ethan Zohn

Again, still early, but these seem reasonable to me. Even pre-season, Domenick talked about wanting to play like Tony. The other comparisons look pretty good as well. We shall see if Brendan can continue to keep up with the Rob-father.

That’s it for this week. If you have questions, hit me up on Twitter, @seanfalconer.

Week 2

Below are the player’s ranked by highest probability to win after three episodes.

  1. Domenick Abbate
  2. Stephanie Johnson
  3. Brendan Shapiro
  4. Michael Yerger
  5. Wendell Holland
  6. Kellyn Bechtold
  7. Sebastian Noel
  8. Bradley Kleihege
  9. Laurel Johnson
  10. Desiree Afuye
  11. Chelsea Townsend
  12. Libby Vincek
  13. Angela Perkins
  14. Chris Noble
  15. Donathan Hurley
  16. Jenna Bowman
  17. James Lim

Quick Thoughts:

The top 5 this week is the same top 5 as in the first week, however Kellyn jumped up a couple of spots to number 6. Not bad considering her tribe hasn’t gone to tribal yet.

Chris took a big hit this week. His codings for aggression and ego were strong this week, which is likely why.

The big surprise for me is that James Lim is in last at the moment. Looking at his behaviors, he is far above the average for being analytical, diplomatic and perceptive, but he also has strong representations for morality and weakness (likely due to the episode 1 challenge). This combination is not a winning one according to the model.

We saw a similar thing happen with Chrissy last season where her early challenge performance hurt her early win probabilities. As the episodes continue, his codings for weakness will likely level out and his rankings will change.

Most Likely Top 3:

The model I am using is trained to predict who is the most likely to win, however, that is very different than most likely to make the top 3. You can have someone in the final 3 that has really no chance to win. In fact, this is often the case.

For fun, I thought it could be interesting to retrain the model to predict most likely to make the final 3 and see how that changes things.

The results are below:

  1. Domenick Abbate
  2. Brendan Shapiro
  3. Chris Noble
  4. Stephanie Johnson
  5. Wendell Holland
  6. Michael Yerger
  7. Kellyn Bechtold
  8. Libby Vincek
  9. Sebastian Noel
  10. Jenna Bowman
  11. Donathan Hurley
  12. Bradley Kleihege
  13. Angela Perkins
  14. Laurel Johnson
  15. Desiree Afuye
  16. Chelsea Townsend
  17. James Lim

The biggest change here is that Chris pops back up to the top 5 when we look at final 3 probabilities. Could Chris be our zero vote finalist this year? Maybe, it’s still early, so hard to say for sure.

That’s it for this week. If you have questions, hit me up on Twitter, @seanfalconer.

Week 3

Below are the player’s ranked by highest probability to win after three weeks.

  1. Michael Yerger
  2. Domenick Abbate
  3. Wendell Holland
  4. Stephanie Johnson
  5. Kellyn Bechtold
  6. Laurel Johnson
  7. Bradley Kleihege
  8. Desiree Afuye
  9. Libby Vincek
  10. Chelsea Townsend
  11. Chris Noble
  12. Donathan Hurley
  13. Angela Perkins
  14. James Lim
  15. Sebastian Noel
  16. Jenna Bowman

Quick Thoughts:

Brendan is our first casualty of the predicted top 5 through the first four hours of Survivor. He is this season’s Ally.

With the tribe mix up, we ended up with quite a bit of shifting in our positioning. Laurel moved up 3 spots to 6th, Michael jumped up into first, and with a weak episode from Stephanie, she dropped to 4th.

I think given the strength of this season’s cast, we will continue to have some mix ups for a while, but I think it’s fairly safe to say that the winner is somewhere in the top 6. There’s a pretty big probability drop off between Kellyn and Laurel and then a huge drop between Laurel and everyone else.

Most Likely Top 3:

I re-ran the contestants against the trained model for predicting top 3 again this week.

The results are below:

  1. Domenick Abbate
  2. Michael Yerger
  3. Chris Noble
  4. Kellyn Bechtold
  5. Wendell Holland
  6. Stephanie Johnson
  7. Sebastian Noel
  8. Donathan Hurley
  9. Laurel Johnson
  10. Libby Vincek
  11. Jenna Bowman
  12. Angela Perkins
  13. Desiree Afuye
  14. Chelsea Townsend
  15. Bradley Kleihege
  16. James Lim

Domenick, Michael and Chris would be a pretty crazy final three. It seems too good to be true.

Besides Chris jumping from a ranking of 11th for win probability to 3rd for making the finals probability, a couple of other big movers are Sebastian from 15th to 7th and Donathan from 12th to 8th.

Chris, Sebastian and Donathan are entertaining and give some great sound bites, but I agree with the model’s prediction that they have low win credibility.

That’s it for this week. If you have questions, hit me up on Twitter, @seanfalconer.

Week 4:

Below are the player’s ranked by highest probability to win after four weeks.

  1. Michael Yerger
  2. Domenick Abbate
  3. Chris Noble
  4. Wendell Holland
  5. Laurel Johnson
  6. Kellyn Bechtold
  7. Bradley Kleihege
  8. Desiree Afuye
  9. Libby Vincek
  10. Donathan Hurley
  11. James Lim
  12. Angela Perkins
  13. Chelsea Townsend
  14. Jenna Bowman
  15. Sebastian Noel

Quick Thoughts:

Some HUGE movement by Chris this week. He went from 11th last week to 3rd this week.

Comparing his behaviors from this week to the prior week, he had his biggest positive shifts in empathy and athleticism, while his level of aggression went down. He’s such a big presence on the season, I can see him bouncing around quite a bit. He has his shining moments like his conversation with Donathan but he also has some pretty rough moments.

We lost Stephanie, who we had at 4th last week. That’s back to back weeks where someone in the top 5 from the rankings dropped out. This may turn out to be a tough season to nail down.

Who’s playing like who?

Now that we are 4 weeks into the season, I wanted to revisit our player comparisons.

I did this by taking each player from season 36 and calculating the dot product between their behavioral vector and all the winners. The combination with the highest dot product is the winner they are most similar to.

The results for the top 6 are below:

  1. Michael Yerger most similar to J.T. Thomas
  2. Domenick Abbate most similar to Todd Herzog
  3. Chris Noble most similar to Tom Westman
  4. Wendell Holland most similar to Ethan Zohn
  5. Laurel Johnson most similar to Kim Spradlin
  6. Kellyn Bechtold most similar to Todd Herzog
  7. Bradley Kleihege most similar to Richard Hatch

In the first week, Domenick was most similar to Tony, but now he’s most similar to Todd. He is definitely less aggressive so far than Tony, so I can see why that comparison has dropped off.

A couple of funny comparisons not shown here, is Sebastian is most similar to Fabio and poor Purple Chelsea is most similar to Bob Crowley. I also had James most similar to Boston Rob and Jenna most similar to Amber. Perhaps a potential showmance? šŸ™‚

Well, that’s it for this week. Cheers!

Week 5:

Below are the player’s ranked by highest probability to win after five weeks.

  1. Michael Yerger
  2. Domenick Abbate
  3. Wendell Holland
  4. Chris Noble
  5. Kellyn Bechtold
  6. Laurel Johnson
  7. Bradley Kleihege
  8. Desiree Afuye
  9. Donathan Hurley
  10. Libby Vincek
  11. Angela Perkins
  12. Chelsea Townsend
  13. Jenna Bowman
  14. Sebastian Noel

Quick Thoughts:

For the third week and a row, Michael and Domenick are number 1 and 2. The top 6 are pretty locked in place right now. Last season the probabilities of the front pack were very close and way ahead of everyone else. This season, there’s more variance in the top 6. Michael and Dom are quite close, and then there’s some drop to Wendell, and then a significant drop down through Chris, Kellyn and Laurel. Everyone else has a negligible probability of winning.

It appears that the AI is not a fan of a showmance as Jenna and Sebastian are bringing up the rear for the second week in a row.

That’s all for this week.

Week 6:

Below are the player’s ranked by highest probability to win after six weeks of episodes.

  1. Michael Yerger
  2. Domenick Abbate
  3. Wendell Holland
  4. Laurel Johnson
  5. Chris Noble
  6. Kellyn Bechtold
  7. Donathan Hurley
  8. Desiree Afuye
  9. Libby Vincek
  10. Chelsea Townsend
  11. Angela Perkins
  12. Jenna Bowman
  13. Sebastian Noel

Quick Thoughts:

Michael and Domenick continue to hold onto the top two spots, while Laurel moved up ahead of Chris.

My prediction is Domenick rallies to take out Chris either through numbers or the use of one of his advantages, then everyone turns on him to get him out. Michael seems too big a threat to not take out, leaving room for Wendell or Laurel to march their way to victory. At least, that’s my fan fiction version of the future :-).

What do you think?

Week 7:

Below are the player’s ranked by highest probability to win after seven weeks of episodes.

  1. Michael Yerger
  2. Domenick Abbate
  3. Wendell Holland
  4. Laurel Johnson
  5. Kellyn Bechtold
  6. Libby Vincek
  7. Desiree Afuye
  8. Donathan Hurley
  9. Jenna Bowman
  10. Chelsea Townsend
  11. Sebastian Noel
  12. Angela Perkins

Quick Thoughts:

Well, the fan fiction that I wrote about last week got at least the first part correct with Chris being knocked out by Domenick. Otherwise, the top winner picks have continued to be locked in place. At this point, Michael, Dom and Wendell all have basically the same win probability.

One interesting thing to note is Libby’s win probability continues to climb. Up until two episodes ago, she really had no win equity, but it has started to rise the last two episodes. Still a very unlikely candidate, but interesting to make a note of.

Who’s playing like who?

I decided to revisit our player comparisons again.

I did this by taking each player from season 36 and calculating the dot product between their behavioral vector and all the winners. The combination with the highest dot product is the winner they are most similar to.

The results for the top 6 are below:

  1. Michael Yerger most similar to J.T. Thomas
  2. Domenick Abbate most similar to Rob Mariano
  3. Wendell Holland most similar to Ethan Zohn
  4. Laurel Johnson most similar to Denise Stapley
  5. Kellyn Bechtold most similar to Natalie Anderson
  6. Libby Vincek most similar to J.T. Thomas
  7. Desiree Afuye most similar to Fabio Birza
  8. Donathan Hurley most similar to Fabio Birza
  9. Jenna Bowman most similar to Amber Brkich
  10. Chelsea Townsend most similar to Natalie Anderson
  11. Sebastian Noel most similar to Fabio Birza
  12. Angela Perkins most similar to Natalie Anderson

Lot’s of Fabio comparisons this time around. Dom switched from Todd to Boston Rob, both strategic and dominate players in their winning seasons. I see Wendell as more strategic than Ethan, but maybe now that Chris is out of the picture, Wendell will get more time to show that side of himself.

What do you think? Let me know on Twitter.

Week 8:

Below are the player’s ranked by highest probability to win after eight weeks of episodes.

  1. Michael Yerger
  2. Wendell Holland
  3. Domenick Abbate
  4. Laurel Johnson
  5. Kellyn Bechtold
  6. Desiree Afuye
  7. Donathan Hurley
  8. Chelsea Townsend
  9. Sebastian Noel
  10. Jenna Bowman
  11. Angela Perkins

Quick Thoughts:

After 5 weeks of ranking Dom with the second highest probability to win, Wendell overtook that spot this week. However, the probability differences between the top three spots are all very small.

The next grouping with any real chance is Laurel and Kellyn. After that, there’s a large drop off to Desiree and past Donathan, no one really has a shot.

I could see Laurel or perhaps Kellyn winning, but I think they’ll need to start taking out the top three players soon. It’s hard for me to see someone winning against Michael, Wendell or Dom if any of those guys make it to the final three.

That’s it for this week.

Week 9:

Below are the player’s ranked by highest probability to win after nine weeks of episodes.

  1. Michael Yerger
  2. Domenick Abbate
  3. Wendell Holland
  4. Laurel Johnson
  5. Kellyn Bechtold
  6. Donathan Hurley
  7. Angela Perkins
  8. Jenna Bowman
  9. Sebastian Noel
  10. Chelsea Townsend

Quick Thoughts:

Minor shake up again in the top 5 with Dom moving back into second ahead of Wendell. Probably the most interesting change is Laurel. Despite her overall position still being 4th in terms of probability to win, her win probability has increased significantly. She’s starting to catch up to the three front runners.

I think it is definitely between those four. It’s hard for me to see a path to the end for Michael just given his threat level. Perhaps if a Kellyn vs Dom/Wendell war starts followed by a Dom vs Wendell war, he can slip in there. I like Laurel a lot, but I think she’s going to need to take out Dom and Wendell to win, but with both of them having influence and idols, I’m not sure how that will happen.

I’m excited to see how things shake out over the next couple of weeks.

Week 10:

Below are the player’s ranked by highest probability to win after ten weeks of episodes.

  1. Domenick Abbate
  2. Wendell Holland
  3. Laurel Johnson
  4. Donathan Hurley
  5. Kellyn Bechtold
  6. Sebastian Noel
  7. Angela Perkins
  8. Chelsea Townsend

Quick Thoughts:

We’ve all been feeling it for weeks now, Michael finally ran out of moves and luck and was knocked off this week. With that shake up, Dom is now our leader. It’s worth mentioning that Michael, Dom and Wendell have had almos the exact same win probabilities since week 4, with a marginal edge to Michael.

At this point, Dom, Wendell and Laurel have almost the same win probabilities. I personally feel like we are looking at a Dom win. With only 8 people left and him with an idol, I don’t think there’s enough people that want to make a move against him and enough time left to knock him out.

Another interesting change this week is Donathan moved up ahead of Kellyn. She’s really looking like the next person out unless she wins immunity.

I think it could make sense for us next season to start incorporating whether you vote for the right person to go home or not as a feature for a player. It seems like a good measurement of power and control in the game.

Week 11:

Below are the player’s ranked by highest probability to win after eleven weeks of episodes.

  1. Domenick Abbate
  2. Wendell Holland
  3. Laurel Johnson
  4. Donathan Hurley
  5. Sebastian Noel
  6. Kellyn Bechtold
  7. Angela Perkins

Quick Thoughts:

Nothing too crazy or surprising this week. Domenick and Wendell continue to dominate the winning probabilities with Laurel having an outside chance. Based on the algorithm, no one else has a shot. Kellyn continues to fall down the rankings.

In my opinion, no one really has a shot to win outside of Dom and Wendell and that’s been the case for weeks. With idols in hand, fake idols, immunity wins and people unwilling to make a move against them, one of the two of them have to win.

I think it’s going to come down to the final 4 challenge and who wins immunity. But even then we could still end up with both in the finals, which would be exciting. I think I’d give the edge to Dom to win, but it’s a very slight edge.

Week 12:

Below are the player’s ranked by highest probability to win after twelve weeks of episodes.

  1. Domenick Abbate
  2. Wendell Holland
  3. Laurel Johnson
  4. Donathan Hurley
  5. Sebastian Noel
  6. Angela Perkins

Quick Thoughts:

It’s now finale week and the rankings are locked in. No movement from last week besides the eviction of Kellyn.

The story of the season has really been Domenick and Wendell running circles around everyone. Pre-merge, there was perhaps some doubt about whether Chris might be able to get Dom out of the game, but ultimately he lacked the skills necessary to get that done.

Bradley with Kellyn could have been a formidable pair to go up against the Dom and Wendell alliance, but in what now looks like one of Dom’s best moves, Bradley was eliminated pre-merge.

The other big story this season in both the rankings and on-screen has been, will underdog Michael be able to somehow slither his way to the end, but he was just too big a juicy target to let him stay in.

It really feels that this season, too many of the people that would have made a move against the Dom and Wendell duo went out pre-merge and then post-merge, the players that were willing to make a move lacked numbers or people willing to try.

In total this season, Domenick has spent 6 weeks at the top of the rankings, 5 in second place and one week in third, with an overall ranking of 1.6. Meanwhile, Wendell has spent 4 weeks in second place, 4 weeks in third, and his lowest placement was 5th near the beginning of the season, giving him an average placement of 3.5.

Laurel is the outside contender. Although her overall ranking for the season is 6.3, starting in week 7, she’s been ranked between 3rd and 4th. The problem is I don’t really see a path to the end for her that doesn’t include at least one of Dom or Wendell and I don’t think anyone can beat either of them.

It feels like the podcast-sphere is predicting Wendell, but our rankings give a slight edge to Domenick. I’m excited to see how it all plays out. Maybe we will all be shocked by a Seabass win :-).

About the author

Sean Falconer

22 Comments

Sean Falconer

Get in touch

I write about programming, developer relations, technology, startup life, occasionally Survivor, and really anything that interests me.