Evaluation of an associative classifier based on position-constrained frequent/closed subtree mining
A real-life event log, taken from a Dutch Academic Hospital, is analyzed using process mining techniques. The log contains events related to treatment and diagnosis steps for patients diagnosed with cancer. Given the heterogeneous nature of these cases, we first demonstrate that it is possible to create more homogeneous subsets of cases (e.g. patients having a particular type of cancer that need to be treated urgently). Such preprocessing is crucial given the variation and variability found in the event log. The discovered homogeneous subsets are analyzed using stateof-the-art process mining approaches. In this paper, we report on the findings discovered using enhanced fuzzy mining and trace alignment. A dedicated preprocessing ProM plug-in was developed for this challenge. The analysis was done using recent, but pre-existing, ProM plug-ins. As the evaluation shows, this approach is able to uncover many interesting findings and could be used to improve the underlying care processes.