Abstract:
|
What if computation was not an obstacle to constructing sparse, accurate models? With new optimization-based techniques for interpretable machine learning, it is not as much of an obstacle as before. I will discuss construction of optimal rule lists (decision trees), scoring systems, and matching algorithms. I will show that it is possible to create interpretable models that are just as accurate as black box models used throughout the criminal justice system for predicting recidivism.
|