psm-batchmining

Detecting and mining batching behavior from event log data.

The Batch Miner (BM) is a Java command-line tool for event data. It takes as input data from the Performance Spectrum Miner [1] in CSV format. The Batch Miner partitions traces for each segment (see [1]) into batches following a set of predefined constraints. It then outputs a new event log in CSV format containing batched and non-batched traces for each segment and several tables containing batch and segment statistics, also in CSV format.

Note: Since activity names may contain characters such as ":" or "/" and these are used as filenames for the output logs, these are all converted to "_" or deleted in the process. This has been successfully implemented for a selected set of logs (BPI11-15, BPI17, BPI19, Hospital Billing and Road Fine) and compatibility with other logs is therefore uncertain.

Prerequisites

• Java 10 (compatibility with older versions of Java is not tested and therefore uncertain).

• Maven: the tool needs to be built as a Maven project since it uses two external dependencies (OpenCSV and RxJava2).

• A directory containing output data of the PSM: this output contains multiple CSV files.

• A directory containing two folders named "Statistics" and "Logs". These directories will be used to store the output produced by the BM.

Note: Output data of the PSM can be obtained by clicking the "Export..." button on the bottom right when using the PSM. This output data also contains a file called "max.csv", which needs to be deleted.

User input

The following parameters need to be specified in the main method of BatchMiner.java:

1. Path to directory containing PSM data

2. Minimum batch size

3. Option to use a 12h non-FIFO time frame for batch detection (recommended for BPIC'17 log)

4. Path to output directory (containing the folders "Statistics" and "Logs")