Recent searches

No recent searches

Search options

Only available when logged in.
sigmoid.social is one of the many independent Mastodon servers you can use to participate in the fediverse.
A social space for people researching, working with, or just interested in AI!

Administered by:

Server stats:

605
active users

sigmoid.social: AboutProfiles directoryPrivacy policy

Mastodon: AboutGet the appKeyboard shortcutsView source codev4.4.1

#Churn

1 post1 participant0 posts today
*
RememberUsAlways

The difference between #Biden and #Trump US market analysis is night and day.

All four years of #Biden, Republicans blamed #Covid benefits for #inflation and the calls for #recession were daily.

Now, "everything" is fine as the US dollar is weaker than at any time since #September11 and a major rotation in markets forced a double peak, all-time high #churn on the stock market.

#Republicans think the stock market IS the US economy when they "feel" power.
#CNBC
#premarket
#investing

Carl Gold, PhD

is in Nature Magazine Scientific reports! Its an interesting study looking at SMOTE and a variety of different classifiers.

TBH I'm taking the conclusions with a grain of salt because (1) They only looked at a single Kaggle dataset; (2) The dataset was not really very imbalanced, with a 20% churn rate; (3) they don't address the calibration consequences of using an oversampling technique like SMOTE.

Still nice that someone published about churn in a journal.

nature.com/articles/s41598-025

Carl Gold, PhD

Recurly published their 2025 State of (I missed this back in Feb) - features industry and a data on re-signups after churn: "Return acquisition percentage" for % of new signups that were previously subscribed. They also say: "Artificial intelligence (AI) has become indispensable. AI enables businesses to analyze subscriber behaviors, predict churn risks, and automate recovery efforts." I always said, Fight With Data!
go.recurly.com/2025-state-of-s

Carl Gold, PhD

comes to !

" businesses are maturing... they face a challenge that publishers have grappled with for years: churn." - Sara Guaglione

Fortunately for them, fighting churn for Podcasters is the same as for everybody else, and the techniques are already well known.

Unfortunately there are no representative in this article - that would have been really interesting.

digiday.com/media/how-podcaste

Carl Gold, PhD

Anntena Web TV has released a new "State of Report". highlights: Churn has "settled" down after climbing since 2022 - 2024 Weighted average monthly churn is 5%. BUT churn is less than 3% when including re-subscribers! Churn & Return lives!

Continued thread
Carl Gold, PhD

5. The datasets have very few features (15-20). Real datasets are created with 50-250 features.

6. The authors don't cite Fighting Churn With Data, the only textbook totally about churn and data science. 😉

So thats what I would have said if I had been a reviewer at the journal! TLDR: Don't use for churn. They take much longer to train/predict and are less reliable.

Continued thread
Carl Gold, PhD

2. models should *not* be evaluated with precision/recall but rather with AUC: True/false churn predictions are NEVER used, but rather risk rankings. (Always use predict_proba for churn, never predict.) Importantly, the precision/recall metrics they show in their results will be sensitive to the thresholds which are not detailed, and thats a tricky issue for imbalanced data. This is another reason not to believe the supposed accuracy improvement.

Carl Gold, PhD

I appreciate the effort in this article on customer prediction in Nature Scientific Reports - and I must give some serious criticism:

1. Gradient boosting ( or LGBM) is state of the art for practitioners in the real world. I don't believe they put much effort into tuning their benchmark models, so I don't believe the claims of higher accuracy. ( researchers always do this - they slave away tuning their preferred model, and use their benchmarks "as is" without tuning.)

TrendingLive feeds
About