Santa Predictive Analytics

Santa’s Chief Data Elfficer contacted Squark over Telegram yesterday.  Santa’s Workshop needed help identifying who was naughty or nice this year.  Santa’s list keeps growing and the elves are collecting more data than ever through the IoT devices, like Elf on the Shelf. He’s also now subject to NPGDPR* and SWPA**.  It was a perfect challenge for Squark’s secure, compliant, and highly scalable no-code AI platform.  We asked Mrs. Claus for permission to share a bit of what we did and the results.  She opted into data sharing. 

Here’s the story of how Squark saved Christmas. 

The Chief Data Elfficer gave us data about kids who Santa previously deemed Naughty or Nice.  Santa’s training data has these features (column names):

We loved this data set. Not only did it give us a peek into Santa’s decisioning process, but we knew Squark’s AI and decision intelligence could help.  The full data set looked something like this:

Santa loves REST API‘s so we gave his data team access. Squark connected to, cleaned, prepared, and engineered, selected, and automodeled Santa’s naughty or nice data. And then Squark built thousands of models using HPO and NAS in minutes.  The best predictive results were from our Stacked Ensemble, which are summarized by our NLG below.

The confusion matrix showed the results of our validation and Santa-folds cross-validation: 

Squark generated scores predicting if each kid was naughty or nice.  Looking good!

Santa was a bit skeptical of machine learning predictions for such a critical task.  We mentioned how these scores should augment his own judgment.  AI can’t replace Santa, but it sure can help him and his little helpers make decisions across billions of data points about people.  In fact, when we showed the Chief Data Elfficer and Santa our AI explanations via our Explainable AI, we know they loved the no-code too.  The elves, in fact, were ushered by Mrs. Claus to the Zoom where we shared what was contributing to kids being judged naughty and nice since they started collecting this data:

Turns our Santa judges nice kids based on getting good grades, the # of books read, doing chores, playing sports, and how much time they spend doing homework each year.  And because of Shapley value symmetry, we also knew what was contributing to naughty behavior.

Turns out Santa doesn’t put much weight on singing Christmas Carols, where you put your stocking, and interestingly on how many mins per day the kids spend on homework.    Total time spent on homework is way more important to Santa’s nice list than average daily minutes spent on homework! 

Santa even asked what why our AI was predicting what is was predicting for each kid.  So we showed him Mikey from Boston, Massachusetts, USA who has been nice.  He checked his notes.  Santa-folds cross validation was correct.  Santa agrees with our AI predictions and explanations! 

Mikey is stoked because his good grades are certainly helping him earn a spot on the nice list this year.

We just finished this machine learning analysis and sent Santa’s Chief Data Elfficer the scores. Santa now wants us to work on a toy recommendation model, so we gotta go. But leave a comment after you check your tree this weekend and let us know how we did when you unwrap your presents (or coal). 



*NPGDPR = North Pole General Data Privacy Requirements
**SWPA = Santa’s Workshop Privacy Agreement

Recent Posts

navigating ai
Predictive AI vs Generative

Squark is a no-code AI as a Service platform that helps data-literate business users make better decisions with their data. Squark is used across a variety of industries & use cases to uncover AI-driven insights from tabular and textual data, prioritize decisions, and take informed action. The Squark platform is designed to be easy to use, accurate, scalable, and secure.

Copyright © 2023 Squark. All Rights Reserved | Privacy Policy