While not “data science” per se (although there could be arguments for and against), in 1992 a kindred predictive study interviewed newlywed couples and identified variables that predicted divorce with 93.6% accuracy (after only five minutes of questioning).
While this study didn’t have a very big data set, the results were statistically significant with the researchers identifying all seven divorced couples out of a total 56. You might have read about this in Malcolm Gladwell’s Blink. There are a lot of related studies and we felt this was a fun example of pattern recognition as applied to an interesting social issue.