To see my presentation to this problem set, go to my Google Drive Slides.
From the Google Doc
Thank you for participating in this project! The main objective of this exercise is to display your analytical and engineering abilities. Please see the instructions below:
We included real borrower data in the email we sent you as a zip file. The zip file contains SMS logs, call logs and contact lists for 200 users who repaid their loan, and for 200 users who defaulted on their loan (400 users in total). The list of users with repaid/defaulted status is in the user_status.csv files within the zip file. For each user and device, the data are organized by:
[user_id]/[device_id]/call_log/*- These contain records of calls the user has made[user_id]/[device_id]/contact_list/*- These contain the contact list on the device[user_id]/[device_id]/sms_log/*- These contain user’s actual SMS messages
Use the raw data in the zip file and the repayment performance of the borrowers (whether they repaid or defaulted) to generate insights that will guide future decisions and underwriting criteria.
Some examples of possible insights might be:
- Users with more than 50 contacts are less likely to default
- The presence of certain types of SMSes are predictive of full repayment
- Making many calls late at night is a sign of likelihood to default
Outline your findings in a simple presentation (a document is fine) that we will review as a group. We do not expect you to find perfectly predictive insights. We realize that drawing definitive conclusions from such a small sample will be impossible.
Please feel free to ask questions while you work on this project. Ping Sarah or add any questions you may have to the bottom of this document. We will respond ASAP so you aren’t blocked.
Questions from Mitchell: As requested by Nick, I'll address questions in the Google Doc