Enhancing model quality and scalability for mining business processes with invisible tasks in non-free choice

Abstract

At present, business processes are growing rapidly, resulting in various types of activity relationships and big event logs. Discovering invisible tasks and invisible tasks in non-free choice is challenging. α$ mines invisible prime tasks in non-free choice based on pairs of events, so it consumes considerable processing time. In addition, the invisible tasks formation by α $ is limited to skip, switch, and redo conditions. This study proposes a graph-based algorithm named Graph Advanced Invisible Task in Non-free choice (GAITN) to form invisible tasks in non-free choice for stacked branching relationships condition and handle large event logs. GAITN partitions the event log and creates rules for merging the partitions to scale up the volume of discoverable events. Then, GAITN utilises rules of previous graph-based process mining algorithm to visualises branching relationships (XOR, OR, AND) and creates rules of mining invisible tasks in non-free choice based on obtained branching relationships. This study compared the performance of GAITN with that of Graph Invisible Task (GIT), α $, and Fodina and found that GAITN produces process models with better fitness, precision, generalisation, and simplicity measure based on higher number of events. GAITN significantly improves the quality of process model and scalability of process mining algorithm.