Improving Declarative Process Mining with a Priori Noise Filtering

In this paper, we report the results of an exploratory study into the efficacy of noise filtering in improving the accuracy of declarative process mining. We apply the double-granularity mixed-dependency filtering algorithm as introduced by [9], to the DisCoveR declarative miner [1], and parameter optimise it to only perform coarse-grained filtering. However, while noise filtering appears promising on the surface, one might worry that the outlier behaviour allowed by declarative models may be wrongly classified as noise and removed. To test the efficacy of noise filtering from both perspectives, we applied DisCoveR with noise filtering to two data sets: the process log collection used in the PDC2020 process discovery contest, emulating “real-world” scenarios; and a synthetic set of logs known to exhibit (non-noise) outlier behaviour. We find that on the contest data sets, noise filtering significantly increases accuracy (on average 23% points), obtaining exploratory evidence that noise filtering may improve declarative miner performance; on the synthetic logs we showcase examples where noise is filtered, while outlier behaviour remains untouched.