On Thursday April 23, 2015 I was invited to present my loan funding predictor project for Kiva.org at the Data for Good Meetup organized by the Brussels Data Science Community.
Here is the link to the 27.5 minute video of the presentation, available exclusively to members of the Brussels Data Science Community. To gain access, simply register first via your Facebook, LinkedIn, Twitter or Google Plus account.
If you prefer not to register or are just short on time, here’s the slide deck:
Postscript (May 14, 2015)
Ultimately, and with hindsight, all four soon-to-expire example loans (1, 2, 3, 4) from this presentation did get fully funded, whereas the model predicted that only one of them would. Does this mean the model is flawed? Not necessarily.
As explained, the prediction is a priori: it does not take into account any effect introduced by, to name one thing, the prominence of a loan on Kiva’s web page with loans that are about to expire. The mere act of highlighting a loan on this page greatly increases the probability for a potential lender visiting that page to actually contribute to the loan.
It would be interesting to know which algorithm the Kiva website builders employ to drive this ranked list. If the algorithm is random, different viewers see different loan proposals; in that case, influence should be minimal, since multiple lenders are needed to fully fund a loan. If not, the act of consistent highlighting obviously steers lending behavior, favoring some loan proposals over others.
The really interesting question, now, is whether the a priori prediction offered by the model should play a role in highlighting soon-to-expire loans (and/or other types of featured loans). Highlighting loans with higher predicted a priori probabilities should increase productivity of money in the system: that’s a global benefit. The other, ethical side of the medal, is whether such a global optimization is fair to loan proposals with lower predicted scores.
As discussed during the Q&A session after the presentation, it is up to Kiva to decide on the right trade-off between individual fairness and global productivity of money.