User engagement and predicting churn of Julius Forum

Hello to everyone,
Introduction : I have made this churn prediction model based the dataset available into the users tab of this forum. I have taken top 42 users data for this from the leaderboard. Lets check the dataset first from here : julius_users - Google Sheets

This dataset appears to be structured for analyzing user engagement and predicting churn based on various activity metrics.

Problem I try to solve from this analysis:

  1. The retention model which tells me the likelihood of an user to return into julius community. Please keep in mind that : Here a user who has visited more than 5 days is considered returning .
  2. To predict user activity based (high,low,medium) on the followings :
    Total activity = No_of_Posts_Read and Replies_Posted and Topics_Created (assuming higher values indicate higher activity) and topics_viewed and no_of_posts_read
  3. To Predict User Churn (likelihood to stop using the forum)

Dateset Heads :

Username: Unique identifier for the user.

Love_react_received: Number of “love” reactions the user has received on their posts or comments.

Love_react_given: Number of “love” reactions the user has given to other users’ posts or comments.

Topics_Created: Number of topics (discussions, threads) the user has created.

Replies_Posted: Number of replies the user has posted to existing topics.

Topics_Viewed: Number of topics the user has viewed (regardless of participation).

No_of_Posts_Read: Number of posts the user has read (might include topics viewed and replies read).

No_of_Days_Visited: Number of days the user has visited the website.

Lets get started

Step 1:

First I have build a retention model: Here a user who has visited more than 5 days is considered returning . Julius has developed a logistic regression model achieved a good accuracy :

Then I asked julius to make a column entitled as user retention and predict YES if the user is returning and No if the user is not returning and export the data. Julius has made this and this is how it looks like

Step 2:
Next I have asked the julius to build a activity table of the users based on the following prompts
Can you predict user activity level based on the followings - Total activity: No_of_Posts_Read and Replies_Posted and Topics_Created (assuming higher values indicate higher activity) and topics_viewed and no_of_posts_read?
Julius has done is successfully and here is the result :

Julius has put a weighted score and based on the score it has classified an user into high- medium-low active user which is now clearly shows the user status in the forum.
The summary :

Distribution of the ‘Activity_Level’:

  • Low: 14 users
  • Medium: 13 users
  • High: 14 users
    This categorization helps in understanding the activity patterns of users based on their interactions on the platform.

Step 3: Now I asked julius to reorder the users based on their activity score from highest to lowest and julius has done it also, if implemented from the leaderboard most active users will be in top and vice versa :

Step 4: Now finally I will do the churn prediction(likelihood to terminate using the forum)
I asked julius for churn prediction based on the followings :

  • “Yes” indicates that the model predicts this user is likely to churn, meaning they are at a higher risk of stopping their use of the forum based on the features analyzed (such as ‘Activity_Score’ and ‘No_of_Days_Visited’)
  • “No” suggests that the model predicts this user is unlikely to churn, meaning they are expected to continue using the forum.

    *** Red points represent users predicted to churn.**
    *** Green points represent users predicted not to churn.**

**Here I got to know some interesting insight. Users scoring from 0-100 are most likely to churn since they have very lower level of visiting days. But one user who has a activity score between 50-60, but has a higher number of days visited are not likely to churn.

On the other hand users who have very high activity score and also have frequent visits to the forum is most likely not to churn.

My entire convestaion with Julus is here,
By the way Julus made some mistakes when I asked him to make a retention graph, it was showing opposite.

Hope this helps othes if anyone wants to do churn prediction. User churn prediction is a technique used by businesses to identify customers who are at high risk of cancelling their subscription or abandoning their account. It basically helps you figure out who is likely to churn (stop using your product or service).

Cite me or give a love reach which inspires me to post more like this. Any mistake I made is not intentional and please correct me.