Predicting-User-Adoption

Task:

This was a take-home assessment from psuedo-company "Relax, Inc." Defining an "adopted user" as a user who has logged into the product on three separate days in at least one seven­day period, my task was to identify which factors predict future user adoption.

Data:

Two .csv files containing the following information:

1. A user table with data on 12,000 users who signed up for the product in the past 2 years
2. A usage summary table with a row for each day that a user logged into the product

Approach:

Adopted users were defined and labeled in the users table based on entries in the logins table. I derived some additional features, including the number of users in an organization, how many organization users were active, whether a user was invited by an active user, how many invitations a user had sent to others who signed up, and how many of their invitees were now adopted users. I fit the scaled & encoded data to an XGBoosted Tree Classifier and then found the SHAP values associated with each feature to interpret their influence on the results.

Results:

Model performance:

AUROC = 0.63

Factors which predict future user adoption, in order of importance:

1. Belonging to an organization with a higher percentage of adopted users
2. Being invited by an adopted user
3. Having an @hotmail.com email address
4. Signing up through Google Authentication or through a guest invite
5. Being opted in to the mailing list
6. Having a @gmail.com email address

Factors which were found to predict a user not becoming adopted:

1. Creating an account for personal projects
2. Having an @yahoo.com email address
3. Having a higher number of users in their organization
4. Being invited through their organization

SHAP Values of Features