Process mining enables organizations to discover, monitor, and analyze their work processes based on data. A fundamental requirement for initiating a process mining project is the availability of an event log, which is not always readily available. In such cases, extracting an event log typically involves various time-consuming tasks, such as writing custom structured query language (SQL) scripts to extract relevant data into an event log format from a relational database. In this work, we explore the potential of large language models (LLMs) to support event log extraction for process mining by leveraging LLMs’ ability to produce SQL scripts. We evaluate the effectiveness of LLMs in assisting this process and analyze their performance across a range of scenarios. Despite the inherent non-determinism of LLM outputs, our findings highlight the potential of future LLM-assisted tools in automating event log extraction, particularly when provided with the appropriate domain and data knowledge context. The implementation of such tools could democratize access to process mining by reducing the need for specialized technical expertise for producing relational database query scripts and minimizing manual effort.

Business process management employs process models and event logs to represent the behavior of the information systems under study. Traditional case-centric notions consider the order of activities and events in isolated process instances. The emerging field of object-centric processes challenges this assumption by putting objects in the center. Object-centric process mining and modeling approaches identify the structure of co-evolving data objects that influence the behavior of an information system to provide a comprehensive view of the system behavior. Object-centricity has been investigated independently in process modeling and in process mining, which resulted in the coexistence of seemingly contradictory assumptions and definitions. As a community effort, this research manifesto relates and aligns existing terminologies, definitions, and perspectives to provide a common ground for current and future research in object-centric business process management. Based on the current state of research, we propose a conceptualization that sets process models and event logs in relation to the information system’s behavior and the execution data it generates. The conceptualization aims at aligning different terminologies and, thus, providing a basis to model and analyze behavioral characteristics. Building on this common ground, we identify open research challenges along the most relevant research areas in object-centric process management. For each research area, its current status is investigated and an outline of the most relevant research challenges is presented.

Prescriptive process monitoring aims to derive recommendations for optimizing complex processes. While previous studies have successfully used reinforcement learning techniques to derive actionable policies in business processes, care processes present unique challenges due to their dynamic and multifaceted nature. For example, at any stage of a care process, a multitude of actions is possible. In this study, we follow the Reinforcement Learning (RL) approach and present a general approach that uses event data to build and train Markov decision processes. We proposed three algorithms including one that takes the elapsed time into account when transforming an event log into a semi-Markov decision process. We evaluated the RL approach using an aggression incident data set. Specifically, the goal is to optimize staff member actions when clients are displaying different types of aggressive behavior. The Q-learning and SARSA are used to find optimal policies. Our results showed that the derived policies align closely with current practices while offering alternative options in specific situations. By employing RL in the context of care processes, we contribute to the ongoing efforts to enhance decision-making and efficiency in dynamic and complex environments.

For a range of reasons, organisations collect data on the work behaviour of their employees. However, each data collection technique displays its own unique mix of intrusiveness, information richness, and risks. For the sake of understanding the differences between data collection techniques, we conducted a multiple-case study in a multinational professional services organisation, tracking six participants throughout a workday using non-participant observation, screen recording, and timesheet techniques. This led to 136 hours of data. Our findings show that relying on one data collection technique alone cannot provide a comprehensive and accurate account of activities that are screen-based, offline, or overtime. The collected data also provided an opportunity to investigate the use of process mining for analysing employee behaviour, specifically with respect to the completeness of the collected data. Our study underlines the importance of judiciously selecting data collection techniques, as well as using a sufficiently broad data set to generate reliable insights into employee behaviour.

Robotic Process Automation (RPA) enables subject matter experts to use the graphical user interface as a means to automate and integrate systems. This is a fast method to automate repetitive, mundane tasks. To avoid constructing a software robot from scratch, Task Mining approaches can be used to monitor human behavior through a series of timestamped events, such as mouse clicks and keystrokes. From a so-called User Interface log (UI Log), it is possible to automatically discover the process model behind this behavior. However, when the discovered process model shows different process variants, it is hard to determine what drives a human’s decision to execute one variant over the other. Existing approaches do analyze the UI Log in search for the underlying rules, but neglect what can be seen on the screen. As a result, a major part of the human decision-making remains hidden. To address this gap, this paper describes a Task Mining framework that uses the screenshot of each event in the UI Log as an additional source of information. From such an enriched UI Log, by using image-processing techniques and Machine Learning algorithms, a decision tree is created, which offers a more complete explanation of the human decision-making process. The presented framework can express the decision tree graphically, explicitly identifying which elements in the screenshots are relevant to make the decision. The framework has been evaluated through a case study that involves a process with real-life screenshots. The results indicate a satisfactorily high accuracy of the overall approach, even if a small UI Log is used. The evaluation also identifies challenges for applying the framework in a real-life setting when a high density of interface elements is present.

The field of process mining has evolved from discovering single work processes towards providing broad insights into peoples’ work practices. Existing techniques can be used to analyse such work practices, but this can be problematic if the available data is limited to the use of a single IT system or is not captured at the right level of granularity. We propose the use of a personal informatics technique, called Active Window Tracking (AWT), as a new way of gathering data for mining work practices. In this study, we identify the opportunities that this technique brings through a case study within our research group. In particular, we show how AWT helps to: capture previously-unrecorded work activities, expose the relations between work processes, and navigate between different levels of data granularity. The technique, which allows for generating new data as well as complementing existing data, is a valuable asset for the community when it comes to better understanding people’s work practices across individual systems and processes.

Predictive Process Monitoring aims to forecast the future progress of process instances using historical event data. As predictive process monitoring is increasingly applied in online settings to enable timely interventions, evaluating the performance of the underlying models becomes crucial for ensuring their consistency and reliability over time. This is especially important in high risk business scenarios where incorrect predictions may have severe consequences. However, predictive models are currently usually evaluated using a single, aggregated value or a time-series visualization, which makes it challenging to assess their performance and, specifically, their stability over time. This paper proposes an evaluation framework for assessing the stability of models for online predictive process monitoring. The framework introduces four performance meta-measures: the frequency of significant performance drops, the magnitude of such drops, the recovery rate, and the volatility of performance. To validate this framework, we applied it to two artificial and two real-world event logs. The results demonstrate that these meta-measures facilitate the comparison and selection of predictive models for different risk-taking scenarios. Such insights are of particular value to enhance decision-making in dynamic business environments.

A process discovery algorithm aims to construct a process model that represents the real-world process stored in event data well; it is precise, generalizes the data correctly, and is simple. At the same time, it is reasonable to expect that better quality input event data should lead to constructed process models of better quality. However, existing process discovery algorithms omit the discussion of this relationship between the inputs and outputs and, as it turns out, often do not guarantee it. We demonstrate the latter claim using several quality measures for event data and discovered process models. Consequently, this paper requests for more rigor in the design of process discovery algorithms, including properties that relate the qualities of the inputs and outputs of these algorithms. We present four incremental maturity stages for process discovery algorithms, along with concrete guidelines for formulating relevant properties and experimental validation. We then use these stages to review several state of the art process discovery algorithms to confirm the need to reflect on how we perform algorithmic process discovery.

An important part of healthcare decision making is to understand how certain actions relate to desired and undesired outcomes. One key challenge is to deal with confounding variables, i.e., variables that influence the relation between actions and outcomes. Existing techniques aim to uncover the underlying statistical relations between actions and outcomes, but either do not account for confounding variables or only consider the process or case level instead of the event level. Therefore, this paper proposes a novel relation mining approach for healthcare processes that 1) explicitly accounts for confounding variables at the event level, and 2) transparently communicates the effect of the confounding variables to the user. We demonstrate the applicability and importance of our approach using two evaluation experiments. We use a real-world healthcare dataset to show that the identified relations indeed provide important input for decision making in healthcare processes. We use a synthetic dataset to illustrate the importance of our approach in the general setting of causal model estimation.

Process mining techniques are valuable to gain insights into and help improve (work) processes. Many of these techniques focus on the sequential order in which activities are performed. Few of these techniques consider the statistical relations within processes. In particular, existing techniques do not allow insights into how responses to an event (action) result in desired or undesired outcomes (effects). We propose and formalize the ARE miner, a novel technique that allows us to analyze and understand these action-response-effect patterns. We take a statistical approach to uncover potential dependency relations in these patterns. The goal of this research is to generate processes that are: (1) appropriately represented, and (2) effectively filtered to show meaningful relations. We evaluate the ARE miner in two ways. First, we use an artificial data set to demonstrate the effectiveness of the ARE miner compared to two traditional process-oriented approaches. Second, we apply the ARE miner to a real-world data set from a Dutch healthcare institution. We show that the ARE miner generates comprehensible representations that lead to informative insights into statistical relations between actions, responses, and effects.

Before the findings of a process mining project can be turned into actionable insights or recommendations, it is essential to make sure that the findings are actually valid. Therefore, the evaluation of the findings is a crucial part of a successful process mining project. Current process mining methodologies, however, fall short in providing actionable support to perform such an evaluation. This is especially true when domain experts are involved. To close this gap, we performed a literature study considering all process mining case studies published in the last two decades. In total, we identified 244 candidate papers of which we analyzed 80 in depth. Based on this literature study, we found a need for a more systematic approach for qualitative evaluations in process mining projects where domain experts are involved. Therefore, we build on these results to propose six validation strategies, which originate from qualitative research. We believe that this proposal for more rigor in the evaluation phase of process mining projects helps to move the discipline forward.

Process models are widely used for various system analysis and design activities, but it is challenging for stakeholders to understand these complex artifacts. In this work, we focus on the use of dynamic visualization techniques, in particular animation, to help reduce users’ cognitive load when making sense of process models. We built on the principles of the cognitive theory of multimedia learning, cognitive load theory , and cognitive dimensions framework to develop an adaptive animation solution. Our experiments suggested that process model comprehension improves when users of process models are provided with animation features; the effect is moderated by process modeling expertise according to a U -shape. Our study contributes to the field of conceptual modeling by making a strong case for the use of animation to support complex problem-solving tasks. Moreover, our animation solution offers ample opportunities for being integrated into industrial modeling tools.

The success of a process mining project is highly dependent on the quality of the event log data, the degree to which quality issues are detected, and the way they are resolved. The detection and resolution of data quality issues requires a systematic approach that is aware of the organisational context in which event log data is created. To this end, the Odigos framework has been developed in prior work. The focus of this paper is the validation of this framework through semistructured interviews with a range of experts in process mining. The experts confirmed the utility of the framework, provided valuable insights into data quality in practical settings, and suggested enhancements to the Odigos framework.

In this work, we propose FlexHMiner (FH), a three-step approach for the discovery of hierarchal models. We formalize the concept of activity tree and event abstraction, which allows us to be flexible in the ways of computing the process hierarchy. We illustrate this flexibility by proposing three different techniques to discover an activity tree: (1) a fully domain based approach (DK-FH), (2) a random approach (RC-FH), and (3) a fall-back, flat activity tree (F-FH). After obtaining an activity tree, the second step of our approach is to compute the logs for each subprocess using log abstraction and log projection. Finally, FlexHMiner discovers a subprocess model for each subprocess by leveraging the capabilities of existing discovery algorithms. Using the domain-based approach as the gold standard and the flat tree approach as base line, we compare the three ways of discovering an activity tree using seven real-life logs.

Abstract Process mining techniques use event logs as input. When analyzing complex databases, these event logs can be built in many ways. Events need to be grouped into traces corresponding to a case. Different groupings provide different views on the data. Building event logs is usually a time-consuming, manual task. This paper provides a precise view on the case notion on databases, which enables the automatic computation of event logs. Also, it provides a way to assess event log quality, used to rank event logs with respect to their interestingness. The computational cost of building an event log can be avoided by predicting the interestingness of a case notion, before the corresponding event log is computed. This makes it possible to give recommendations to users, so they can focus on the analysis of the most promising process views. Finally, the accuracy of the predictions and the quality of the rankings generated by our unsupervised technique are evaluated in comparison to the existing regression techniques as well as to state-of-the-art learning to rank algorithms from the information retrieval field. The results show that our prediction technique succeeds at discovering interesting event logs and provides valuable recommendations to users about the perspectives on which to focus the efforts during the analysis.

Conformance checking enables organizations to automatically identify compliance violations based on the analysis of observed event data. A crucial requirement for conformance-checking techniques is that observed events can be mapped to normative process models used to specify allowed behavior. Without a mapping, it is not possible to determine if an observed event trace conforms to the specification or not. A considerable problem in this regard is that establishing a mapping between events and process model activities is an inherently uncertain task. Since the use of a particular mapping directly influences the conformance of an event trace to a specification, this uncertainty represents a major issue for conformance checking. To overcome this issue, we introduce a probabilistic conformance-checking technique that can deal with uncertain mappings. Our technique avoids the need to select a single mapping by taking the entire spectrum of possible mappings into account. A quantitative evaluation demonstrates that our technique can be applied on a considerable number of real-world processes where existing conformance-checking techniques fail.

Healthcare can be considerably expensive for both patients and insurance companies. In some cases, high costs in healthcare are an indirect outcome of a low quality of care, for example, when treatments have to be repeated. Unfortunately, identifying the factors that lead to such repetitions is a complex and challenging task. In this paper, we focus on the domain of dental healthcare and develop an approach that can predict treatment repetitions in the context of the implant denture therapy process. The challenges associated with predicting treatment repetitions in this setting are considerable. First, hardly any patient undergoes the exact same series of treatments like another. This results in a high degree of variation in the data. Second, only a few patients experience treatment repetitions. This lead to a highly imbalance in the data. To address these challenges, we develop a prediction technique that particularly exploits the process perspective. What is more, we apply so-called resampling methods to deal with the imbalance in the data. Our resulting model is able to predict treatment repetitions with an AUC value of 0.69.

The image of workflow systems as being context-insensitive technology, hindering rather than supporting people in performing their work may still exist at present. This impression is also raised in the well-known and often cited case study within Establishment Printers. Using this case as a starting point, this paper presents an analysis of more recent workflow implementations to support the view that modern workflow systems are widely applied in the services industry and are considered useful by performers to support their way of working. In cases where the introduction of workflow technology initially disrupted the flow of work, a wide range of configuration options was available to mend such situations. A detailed analysis of a workflow implementation in a Belgian financial organization clearly shows that re-configuration decisions, like a finer step granularity, can transform a pre-structured production-type workflow system into a flexible application allowing and supporting a smooth flow of work.

On the Dutch workflow market, a new and interesting paradigm named "case handling" is emerging. The goal of case handling is to overcome the limitations of existing workflow management systems. By using a data-driven approach combined with implicit routing and carefully avoiding context tunneling, awareness and flexibility are improved. Currently, many organizations are considering case handling systems such as FLOWer (Pallas Athena) rather than the more traditional workflow management systems. This paper provides a critical assessment of this development. The goal is to show the pro's and con's of case handling. Moreover, based on this assessment, an alternative approach using slightly extended workflow management systems is proposed. This approach is being pursued by the Dutch government in a project involving the workflow management system Staffware. Based on our experiences thus far, we provide guidelines for selecting the proper technology.