Scribd
Upload
13 views1 page

Features For Predicting Active Users

The document discusses the key features for predicting user activity, highlighting that account creation time is the most significant factor, along with organization affiliation, marketing participation, and creation source. Out of 12,000 users, only 1,602 (13%) were classified as active, and a boosted random forest model achieved 92% accuracy in predicting active users. Further investigation is suggested to understand the importance of creation time and the impact of different organizations and sources on user activity.

Uploaded by

jayantshrt0
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
13 views1 page

Features For Predicting Active Users

The document discusses the key features for predicting user activity, highlighting that account creation time is the most significant factor, along with organization affiliation, marketing participation, and creation source. Out of 12,000 users, only 1,602 (13%) were classified as active, and a boosted random forest model achieved 92% accuracy in predicting active users. Further investigation is suggested to understand the importance of creation time and the impact of different organizations and sources on user activity.

Uploaded by

jayantshrt0
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd

Features for Predicting Active Users

The most important features in predicting whether users become active is the time they
created their account. However, several other factors are also important, namely: the
organization they belong to, if they were part of a marketing drip, and what their creation source
was. To find this I first determined how many users fit the “active” criteria of having logged in at
least three times in a 1 week period. Of the 12,000 users, only 1602 were active – or about 13%.
Then I used a boosted random forest to determine which features best predicted the active users.
The model had 92% accuracy, though with tuning I’m sure that could be increased.
With the overall active rate being so low, it seems that further investigation would be
warranted into what about the creation time is so important. Did most active users join at certain
times? This with the marketing campaign variable this seems to imply that. Which creation
sources and which organizations tend to produce the most active users. These would be good to
investigate. Additionally, I expected that those who were invited by another user would be more
likely to be active, but that feature had no influence (or at least was redundant with another).
This could be due to me misinterpreting that information. I assumed that the missing data in that
column meant the user was not invited. If that is inaccurate, the additional data could be helpful
in making that feature meaningful.
-Matthew Rytting

You might also like