Coarse-Grained Process Diagnostics: A Method Combining Process Mining and Time Series Analysis

 

Coarse-Grained Process Diagnostics: A Method Combining Process Mining and Time Series Analysis

Student: Firas Gharbi

Title: Coarse-Grained Process Diagnostics: A Method Combining Process Mining and Time Series Analysis

Supervisor: Mahsa Bafrani

1st Examiner: Prof. Wil M.P. van der Aalst

2nd Examiner: Prof. Dr. Stefan Decker

Summary

Most information systems supporting business processes record event data. Process mining
is a set of data-driven techniques that help organizations to understand their processes. The
techniques provide actionable insights to the business owner by using recorded event data
as the source. For example, performance and compliance issues within organizations can
be identified and processes can be improved based on the resulting diagnosis. To evaluate
process performance and identify best practices as well as opportunities for improvement,
business owners need a global perspective on how their processes behave over time. Most
techniques analyze the process at fine-grained levels, i.e., at the instance context. This
can result in missing insights at higher context levels in the analysis and does not provide
results suitable for long-term decision-making. Therefore, this thesis presents a generic
framework for coarse-grained process diagnostics to consider different granularity levels of
processes. The framework uncovers performance and deviation issues in fine-grained analyses,
enhances diagnostics by unifying fine-grained analyses with coarse-grained analyses,
and supports future analysis of processes by predicting the process behavior over time. We
start with fine-grained performance and deviation analysis of processes to identify potential
weaknesses inside the organization over an aggregated fine-grained view. For instance,
a resource with low productivity or an activity with a long execution time over the entire
process time-span. Based on the results, we systematically extract process aspects from
the event data on the basis of coarse-grained process logs. The approach continues with
analyzing the aspects over time by applying multiple time series analysis techniques. The
resulting insights into patterns, relations, dependencies, and causes-and-effects, are used to
diagnose the root cause of the weaknesses. To demonstrate the applicability of our framework,
we apply it to both synthetic and real-world event logs and show that it is capable
of improving the diagnosis by providing new insights at coarse-grained levels.