This blog accompanies our festive “12 Days of Data Science” video series, where we use the familiar structure of the Twelve Days of Christmas song to explain core data science concepts in a fun, accessible way. Each “day” introduces a practical idea used by data scientists in the real world, including six degrees of separation and network analysis, fraud detection using Benford’s Law, k-nearest neighbours, confidence intervals, top-N ranking lists, error detection with ISBN check digits and modulo arithmetic, and time-series forecasting based on seasonal trends. By linking these technical topics to everyday experiences, such as social media recommendations, online shopping at Christmas, detecting fake numbers, or predicting busy retail periods. we show how data science helps analyse patterns, make predictions, and support better decision-making. The series is designed to demystify data science for non-technical audiences while still accurately reflecting how modern analytics, algorithms, and machine learning are used behind the scenes in platforms like YouTube, LinkedIn, and e-commerce systems

Can data science write the perfect Christmas song? In this video, we use real Spotify data, machine learning, and music analysis to uncover what actually makes a festive hit. From Lyrics and sentiment arcs to keys, BPM, and danceability, we build a model using TF-IDF and Elastic Net regression to find out what works and what definitely doesn’t.

Along the way, we battle the Spotify, API, analyse classics from Mariah Carey to Slade, and turn messy real world data into a statistically informed Christmas song… which we then actually write.

You’ve heard this argument a million times. Some people are like, “It’s set at Christmas, there are carols, of course it’s a Christmas movie.” Other people: “It’s just an action film with tinsel, calm down.” So instead of screaming on the internet for the rest of my life, I did the only sensible thing: I built a model.

In this video I’m going to show you how I trained a simple Christmas-movie classifier using soundtrack data and basic movie metadata, and then used it to settle the age-old internet debate so we never have to talk about it again. We’re literally going to turn the Die Hard argument into machine learning.

So, what was the actual goal here? I wasn’t trying to build the biggest, fanciest AI in the world. The question was much pettier than that: can I take the stuff people always bring up in the Die Hard argument and turn it into numbers, train a model on a bunch of movies, and ask it, “Okay, based on everything you’ve seen… what do you think about Die Hard?”