I want to perform "reliable rule learning", i.e. mining a set of rules with a very low number of false negatives. I recently read the paper "Reliable agnostic learning" by Kalai et al. (https://doi.org/10.1016/j.jcss.2011.12.026) and they basically describe what I want: Rules are determined to reliably classify data points, and the reliability is partly reached by allowing "I don't know" as an additional answer. Sadly, their paper is purely theoretical and I could not find a corresponding implementation. Is there an implementation of reliable rule learning or a similar algorithm? At best, open source and Java-based?