Figuring out high options and understanding how they have an effect on prediction outcomes of machine studying fashions with SHAP
This text is one in all a two-part piece documenting my learnings from my Machine Studying Thesis at Spotify. Make sure to additionally take a look at the second article (incoming very quickly!) on how I succeeded in considerably optimizing my mannequin for this analysis.
Two years in the past, I carried out a captivating analysis challenge at Spotify as a part of my Grasp’s Thesis. I discovered a number of helpful ML methods, which I imagine any Information Scientist ought to have of their toolkit. And at the moment, I’m right here to stroll you thru one in all them.
Throughout that point, I spent 6 months attempting to construct a prediction mannequin after which deciphering its inside workings. My objective was to know what made customers glad with their music expertise.
It wasn’t a lot about predicting whether or not a person was completely happy (or not), however somewhat understanding the underlying elements that contributed to their happiness (or lack thereof).
Sounds thrilling, proper? It was! I beloved each little bit of it as a result of I discovered a lot about how ML will be utilized within the context of music and person habits.
(Should you’re within the purposes of ML within the music business, then I extremely suggest trying out this fascinating research led by Spotify’s high consultants. It’s a must-read!)
In tech, analysis initiatives like mine are quite common as a result of a variety of the work revolves round delivering one of the best customized expertise for customers/prospects.
This typically means delving into the human psyche, and ML could be a highly effective device for reaching the not possible — understanding people.
After we mix ML with Psychology and Behavioral Sciences, we get nearer to having an entire image of how people behave.
How?
