While I was looking for a dataset for my previous post, I found one on Married at first sight (MAFS). MAFS is a show where a group of singles agree to marry partners at first sight chosen for them by a team of relationship experts. The experiment goes for 8 weeks, and the couples decide to stay married or get a divorce. MAFS is my favourite reality show and I like to make predictions on which couple stay married or get a divorce before the show ends.
Even though I like to make predictions based on my intuition, I never thought of using past data to predict new ones till I saw the dataset on Kaggle. Then, I decided to build a model to predict the decisions of the new season that premiered on my birthday :D.
Cleaning the dataset:
I found a dataset that was prepared in a way that showed an individual’s decision, not a couple’s. There is nothing wrong with that as that is how the show works.
At the end of the season, the husband and wife make their decision separately. However, I wanted to predict the final decision regardless of the individual’s decision. So I put each couple together in a row.
| Wife’s decision | Husband’s decision | Final decision |
| Yes | Yes | Yes |
| Yes | No | No |
| No | Yes | No |
| No | No | No |
I also don’t see the relevance of the experts’ names in the dataset so I dropped the columns with the names. Their decisions vary and one expert might have a higher success rate but there isn’t enough of that data for this experiment at this point.
There are a lot of occupations, so I made them into categorical variables.
I created 7 occupation categories (A – G)
- A – This category is for professionals who need a degree to practice in their field (e.g Doctors, Lawyers, Architects etc)
- B – This category is technical jobs that usually call for a degree or training (Realtors, Bookkeepers, Sales representatives etc)
- C – This category is for jobs that usually call for high school and/or job-specific training (Make-up artist, air hostesses, receptionists etc)
- D – This category is for labour jobs that usually give on-the-job training (cleaners, maids etc)
- E – This category is for students
- F – This category is for athletes and people in the fitness industry. I created this category because I noticed a trend of experts matching people in the fitness industry together
- G – This category is for managers and entrepreneurs
Also, I included a new column called ‘Marriage Consummated‘ because attraction is a big problem when strangers marry each other. I would like to see if sex plays a huge role in a couple’s decisions. I didn’t watch s07 and 08 but I hope I got the right answers from google.
You can find my clean dataset here. To save, right-click and click on Save As.
Building the model
I built my model in Alteryx using logistic regression and passed in data from seasons 1 to 10 so I could try it out on seasons 11 and 12 participants. Season 1 – 10 have only 34 couples.
The model has a 92% accuracy. It does however has a low R2 value and high p-value. This tells us the model is not very good at making accurate predictions because there is a great deal of unexplained variance. Our sample size is too small.
When I ran the model, it was 50% accurate
Looks like my model is a hopeless romantic, it predicted that 9 of the 10 couples would decide to stay married and in reality, the only couple it didn’t predict would stay together actually did.
I put in the real results from seasons 11 & 12 and built another model to predict the results for the new season.
Results
Since I cannot tell if they would consummate their marriages, I ran the model against two datasets. One where they all consummate, and the other where they don’t.
From the results, it looks like they all stand a chance of staying together if they consummate their marriages.
However, if they don’t consummate their marriages, Rachel and Jose alone do not stand a chance.
Personally, I think three couples might end up getting a divorce, I do agree that the last couple to make it is Rachel & Jose. So they probably really are a bad match but I guess we have to wait and see.
My predictions are:
- Michaela & Zack – Stay Married
- Bao & Johnny – Stay Married
- Myrla & Gil – Get divorced
- Brett & Ryan – Get divorced (I like this couple and I hope they prove me wrong and stay married)
- Rachel & Jose – Get divorced (even when all I saw was a picture, I wasn’t feeling this match)
There is a fan page (MAFSfan on Instagram) I follow, and here are the predictions from fans after learning a little bit about the couples.














