Now that I have two weeks of predicting Strictly scores under my belt, I’m interested in looking at who gets voted into the dance-off. For example, a 2016 article in The Guardian by Julia Carter and Richard McManus reported that “after controlling for where the couple have come in the judges’ scoring, an ethnic minority celebrity is statistically significantly more likely to be in the bottom two and therefore to have received a lower public vote.

Lots of surprises during Week 5 of Strictly Series 16! Not from this week, but still a surprise. To predict Week 5 Strictly results, I retrained the same three models as I used previously for Week 4, adding in the Week 4 results as additional model inputs. Since this represents a 33% increase in available data about the performance of this series’ celebrities, I was hopeful the model predictions would improve.

The judges have their scores! Let’s see how they compare to the Week 4 predictions. celebrity professional dance fan gbr xgbr rfr actual Ashley Roberts Pasha Kovalev Tango 36 32 33 28 32 Charles Venn Karen Clifton Salsa 25 24 24 25 25 Danny John-Jules Amy Dowden Viennese Waltz 29 29 30 27 27 Faye Tozer Giovanni Pernice Rumba 36 30 28 29 29 Graeme Swann Oti Mabuse Jive 30 22 25 23 26 Joe Sugg Dianne Buswell Cha-cha-cha 28 26 28 25 26 Kate Silverton Aljaž Skorjanec Samba 28 25 26 26 20 Katie Piper Gorka Márquez Jive 21 17 17 20 18 Lauren Steadman AJ Pritchard Quickstep 25 23 24 26 25 Dr.

It’s time to predict some scores! Last time, I walked through how I trained three optimized machine learning models to predict Strictly Come Dancing scores, using all available data through Week 3 of the current series, Series 16. All three models performed decently, but the gradient boosting regressor tended to achieve the best accuracy scores in the cross-validation. So, I used that model as the “official” prediction of the week’s scores, though I was interested to see how the other models’ predictions compared.

Now that in the previous post I’ve defined the goal of our model (predict Strictly scores out of 40 for the upcoming week!) and defined teaching and testing inputs and labels from the data, it’s time to build some models! Setting up models I’ve decided to compare the performance of three different regressors. One is a random forest regressor (RandomForestRegressor from sklearn.ensemble) and two are different gradient boosting regressors (GradientBoostingRegressor from sklearn.

So! After getting the data prepared in the previous post, I should be ready to use machine learning and build models? Not quite. First, I need to think a little more about the requirements of the model, how I will use the available data, and how to prepare the data for the model. Model requirements I want to see how well a model can predict the total score a partnership will receive for a given dance.

The first step in my analysis is to collect Strictly data. Fortunately a lot of the hard work has been done for me by dedicated Strictly fans! Anyone who has spent time on Wikipedia will not be surprised to hear the encyclopedic collection of Strictly data available there. There’s not only all of the scores for all 16 series, but also information like the highest and lowest score for each dance, for each series.

Welcome to Strictly Come Data! I’m analyzing the data behind the popular BBC programme Strictly Come Dancing in a series of posts. Some questions I’m interested in: Can I create a model to predict dancers’ scores for upcoming weeks of Strictly? Can I create a model to predict who will be in the dance-off? Stay tuned, and keeeeeep data-ing!