This paper presents a novel discovery approach in process mining methodology that facilitates the incremental derivation of a series of Structured Information Control Net (SICN) process models. Each model is discovered from selectively grouped temporal traces within large-scale and fragmented process log datasets, enabling its effective and efficient application in the life-cycle management of the corresponding original process model. By conceptualizing the framework and implementing its system, the paper establishes a practical foundation for the proposed approach and ensures that gradual discovery is executed while preserving complete control-flow structures across selected process event log trace groups. Grounded in the SICN process modeling methodology, this approach is further supported by the development of a novel gradual rho-Algorithm, an extension of the original rho-Algorithm. The original rho-Algorithm is a validated process mining technique designed to extract SICN process models from datasets of process enactment event logs and histories. The proposed approach is particularly advantageous for managing the life cycles of complex and large-scale business process models, coping with offering both scalability and efficiency. Conclusively, the primary objective of the paper is to discover and visualize a series of SICN process models, each of which is gradually discovered from selectively chosen temporal trace-groups within large-scale and fragmented process log datasets and completely and safely (matched pairing and proper nesting properties) constructed from combinational building blocks derived from four fundamental process patterns, such as linear (sequential), disjunctive (exclusive-OR), conjunctive (parallel-AND), and repetitive (iterative-LOOP). To validate the conceptual robustness and functional accuracy of the proposed approach and system, three experiments are conducted. These experiments assess the approach using synthetic non-noisy datasets as well as real-world datasets, all of which are publicly available through the 4TU.ResearchData repository. The datasets utilized adhere to the standardized XES (eXtensible Event Stream) data format established by IEEE.

In order to effectively do reengineering a (workflow or business) process model as a lifecycle-managing activity, it is important to secure an elaborate process mining system that is able to discover the enacted process model from an enactment event log dataset of a corresponding process model. In evaluating the process mining approaches and systems, there are two fundamental research issues; One is the discovery accuracy issue and the other is the discovery adequacy issue. The former is about the syntactical quality of the discovered process model, while the later is about the semantical quality of the discovered process model. In this paper, we initiate an innovative process mining approach, which is named as the explainable process mining approach, as a silver bullet that can be a decisive solution for both of the research issues at the same time. We also validate its practical feasibility through an operational experiment on an exemplary dataset available from the 4TU Center for Research Data. Ultimately, through the explainable process mining approach and system, it ought to be possible for the modelers and the mining engineers to effectively explain the reasons why the structural formations of control-flow gateway-activities of the discovered process models are shaped as well as stemmed from in terms of fulfilling the process reengineering works in the process lifecycle management activities.

This paper presents a novel big-data generation framework, COME-HVLM (Cognitive Object–Motion–Environment Hybrid Vision–Language Model), which integrates multiple deep learning–based vision–language architectures into a unified hybrid model. Built upon the foundations of established and advanced VLMs—such as YOLO, BLIP, CLIP, and VILA—COME-HVLM is designed to automatically detect and interpret active contextual elements including objects, motions, and environmental features within CCTV surveillance video streams. Operating primarily in edge-computing environments of smart city surveillance systems, the framework systematically extracts and encodes these contextual elements—termed CCTV-Surveillance Active Contexts—into a structured representation format known as COME-Code (Contextual Object–Motion–Environment Code), expressed in JSON. Furthermore, COME-HVLM supports the augmentation of these COME-Code datasets with additional semantic attributes and descriptive properties, thereby enabling enriched and machine-processable contextual representations for large-scale CCTV video analytics and intelligent surveillance applications.

Managing the CCTV recorded footage takes advantage of not only hardware resources but also labor costs in monitoring and searching content through vast amounts of video clips generated every minute from CCTV cameras. Many studies and practices were introduced to deal with this matter. However, several challenges still need to be solved when implementing a video retrieval system. This study explores the applications of deep learning and graphs to create indexed data in order to implement functions in a video retrieval system. In particular, in this paper, we proposed an approach using graph structure to index the content in CCTV videos after using deep learning algorithms to contextualize detected objects in frames from these videos. Furthermore, we extend the approach to manage all the CCTV videos using graph management approach. This strategy, on the one hand, takes advantage of graph structure, providing an intuitive view for the CCTV management centers in managing and searching content from the managed cameras. On the other hand, it offers easy and efficient retrieval of video content.

Deep learning architectures extract insights from extensive datasets, thereby enhancing human decision-making with more rational, agile, and precise predictions. Recent research in business process management has increasingly leveraged deep learning to support predictive process monitoring-encompassing tasks such as next-activity prediction, remaining time estimation, and workload forecasting. In this paper, we introduce a novel framework designed to forecast subsequent activities in the temporal sequences of individual process instances. Our method integrates a mixture-of-experts mechanism with the Switch-Transformer neural network architecture, and the model is trained from scratch using historical process enactment logs to estimate the probabilities of upcoming activities. Furthermore, the predicted activities are displayed in a temporal graph that preserves their sequential order for each running process instance. This visualization not only clarifies the predictive process monitoring paradigm and its underlying decision-making workflow but also streamlines risk assessment and analysis in business process monitoring operations. To validate the operational efficacy of our approach, we conducted extensive experiments on both synthetic and real-world process event log datasets commonly employed in previous studies. The results demonstrate that our method not only outperforms existing techniques in predictive accuracy but also offers superior usability and practical applicability in real-world settings.

This paper proposes a deep learning model for predicting the next activities in a temporal sequence of activity-events associated with a single process instance. The proposed model is based on a mixture of experts (MoE) mechanism supported by the Switch-Transformer neural network architecture. The deep learning model is trained from scratch on process enactment event log histories. The ultimate goal is to predict the next activities, generate their probabilities of occurrence, and visualize all the predicted next activities in a graphical form of temporal order representing each running case of the corresponding process model. To assess the operational effectiveness of the proposed model, we successfully conducted performance evaluations through a series of experiments on synthetic and real-life process event log datasets, which have been typically and usefully applied in other approaches in previous research works.

In business process life-cycle management and reengineering through process mining, it is crucial for the process mining system to discover structurally safe and complete business process models from process logs. However, most process mining systems typically suffer from discovering inaccurate and unstructured business process models due to various types of anomalous traces hidden in process logs. In this paper, we therefore propose a process log preprocessing approach that purifies a corresponding process log dataset by selecting a group of reasonable process log traces and applying to discover an essential business process model satisfying the structural completeness requirement of matched-pairing and proper-nesting properties. The theoretical basis of the proposed approach is the mathematical graph model of information control net model that is the typical representation of business process models. Finally, we practically demonstrate the conceptual excellence and functional correctness of the proposed preprocessing approach through experimental verifications using real business process enactment log datasets available in the 4TU Center for Research Data.

This study presents a novel, automated approach for the detection of physical and social disorder using contextualized video data from CCTV footage. By leveraging deep learning techniques, this method enhances the precision and scalability of urban disorder surveillance. Traditional methods, such as Systematic Social Observation (SSO) and surveys, face inherent limitations, including observer bias and scalability issues, which reduce their effectiveness in large-scale or long-term applications. To overcome these challenges, this research implements deep learning models, specifically YOLO and BLIP, to extract a wide range of contextual information from video frames, improving the accuracy and scalability of detecting social and physical disorder. The extracted data is categorized using a pre-defined disorder word dictionary, allowing for structured and systematic analysis. The proposed model automates the entire detection process, eliminating the need for human intervention and significantly reducing the time and costs associated with traditional methods. Additionally, the model transforms large-scale video data into structured text, facilitating efficient storage and retrieval. By calculating spatial distances between individuals and hazardous objects, the model also enables the identification of complex social disorder signs that are difficult to detect with conventional video analysis techniques.

This keynote speech is to all intents and purposes introducing a new process mining approach and its implemented system, which are named as eXplainable process mining (XPM) approach and system, respectively. Through the XPM approach, we can selectively amalgamate a group of process event log traces, and through the XPM system, we can ultimately discover a series of complete and safe process models from a process enactment event log dataset. The XPM approach is theoretically based on the typical process modeling methodology of structured information control nets (abbreviated as SICN) and its algorithmic functionality is substantiated by the explainable ρ -Algorithm that is extended from the original ρ -Algorithm [6]. This keynote speech is also explicating the proven process mining algorithm that is able to discover the underlying SICN-oriented process model from a dataset of process enactment event logs and histories. More specifically, the concrete goal of this keynote speech is about an explainable way of discovering the complete structural formation of SICN-oriented process model. Assume that the structural formation is made up of an arbitrary number of combinational building blocks of the primitive process patterns such as linear (sequential), disjunctive (exclusive-OR), conjunctive (parallel-AND), and repetitive (iterative-LOOP) process patterns. Conclusively, it would be emphasized that the XPM approach and system ought to be very effective and well-fitted for managing life-cycles of very large-scale process models deployed in process-aware enterprises and organizations.

Nowadays, with the widespread use of CCTV cameras everywhere, traditional video retrieval systems face significant challenges in managing and retrieving video content from a huge number of CCTV cameras. These systems often have difficulty understanding the semantics of content recorded from cameras and have problems processing large volumes of CCTV video data in real-time. In particular, accurately identifying specific events, objects, or actions is usually time-consuming and may sometimes lead to missing important information. To address these issues, in this paper, we propose an AI-agent-powered video object contextualization and retrieval system that leverages the capabilities of distributed AI agents and edge computing. Our proposed architecture deploys video object contextualization machines on edge devices to perform real-time object detection and classification and then transfer the data to a central server through text-based content. This process extracts rich, semantically meaningful metadata directly from CCTV cameras, bringing frame content into contextual information that we can search using human natural language with the support of NLP agents. By embedding the concept of video object contextualization at the edge, the system significantly reduces data transmission to the central server, enhances processing efficiency, and enables rapid video retrieval. In particular, we present the entire architecture, components, data flows, and implementation of the system in a real-world scenario. The experimental results indicate that the architecture and system are efficient in managing CCTV cameras and supporting traditional CCTV management systems.

In order to effectively do the reengineering of process models within managing their lifecycles, it is important to maintain the qualitative levels of control-flow gateway-activity formations of the discovered process models, as highly as possible, in the process mining time. However, almost all the conventional process mining systems are hardly assured the consistent functionality in discovering the higher level of qualitative formations; Setting aside the quality issue, some of them even suffer from discovering inaccurate formations. Therefore, this paper tries to devise an algorithmic approach of detecting a small group of traces in a process log dataset, which may exert strong effects on degrading the qualitative level of control-flow gateway-activity formations of the discovered process model, because these traces negatively affect to form the structural formations during the process mining time. In detail, this paper formulates the new type of exceptional traces, which is named as structural anomaly-trace type, within a process log dataset, and formally defines a series of related conceptual definitions. Next, an algorithmic approach to automatically detect the structural anomaly-traces is described, and its functional correctness is verified through its implemented process mining system operating with an exemplary dataset of the Customer Summary process logs available from the 4TU Center for Research Data. Finally, this paper validates and discusses the feasibility as well the effectiveness of the proposed approach on the process mining time as well as the process reengineering time of the process lifecycle management scheme in process-aware enterprises and organizations.

In this paper, we introduce a web-based system that leverages the capabilities of the rho(rho)algorithm, which is a Structure Information Control Net (SICN)-oriented process mining algorithm, with open-source platforms, including Django, Graphviz, and Cytoscape, to facilitate the rediscovery and visualization of business process models. Our approach involves discovering SICN-oriented process models from process instances from the IEEE XESformatted process enactment event logs dataset. This discovering process is facilitated by the rho-algorithm, and visualization output is transformed into either a JSON or DOT formatted file, catering to the compatibility requirements of Cytoscape or Graphviz, respectively. The proposed system utilizes the robust Django platform, which enables the creation of a userfriendly web interface. This interface offers a clear, concise, modern, and interactive visualization of the rediscovered business processes, fostering an intuitive exploration experience. The experiment conducted on our proposed web-based process discovery system demonstrates its ability and efficiency showing that the system is a valuable tool for discovering business process models from process event logs. Its development not only contributes to the advancement of process mining but also serves as an educational resource. Readers, students, and practitioners interested in process mining can leverage this system as a completely free process miner to gain hands-on experience in rediscovering and visualizing process models from event logs.

In the research field of the automated process discovery and analysis, the purity of event log datasets ought to be a matter of vital importance to the success of discovering sound and exact process models. Moreover, there exist various types of anomalies that result in the discovery of inaccurate process models from the process enactment event log datasets. A peculiar one out of these anomalies, which is the core challenging issue of this paper, is the temporal activity-sequencing anomaly that critically affects the overall quality of the automated process discovery. This paper explores such event-log anomalies and noises produced by the special type of anomalies inevitably formed in the event-log preprocessing phase of the automated process discovery. More precisely, it implements an algorithmic approach that is able to detect and filter out those anomalies and noises in performing the automated process discovery. The author also carries out a series of experimental analyses by applying the implemented approach to the five datasets of process event logs available in the 4TU Center for Research Data.

Crowd density level evaluates the number of people gathering in a particular area or scene, typically based on videos generated from real-time closed-circuit television (CCTV) cameras. The goal of estimating the levels of a crowd is to provide accurate and reliable information about the crowd size and density, which can be helpful in various situations. Furthermore, detecting and analyzing crowds in real-time is a crucial task for many applications such as monitoring large crowds in public areas, detecting any suspicious activities or potential threats; managing crowds in public places to prevent overcrowding and ensure the safety of individuals; monitoring the flow of people in an event to optimize the use of security officers and facilities. In this paper, we present a deep learning-based architecture and system for detecting levels of crowd density from CCTV cameras. In particular, the system can automatically generate alert information when identifying high-density camera areas, enabling security staff to take corrective measures to manage the crowd and prevent overcrowding before it happens. The proposed system provides a reliable approach for real-time crowd detection, which can be integrated into existing surveillance systems.

In recent years, we can see the presence of CCTV cameras everywhere in developed cities and urban areas. CCTV cameras are widely equipped in public places, buildings, and entertainment areas. Along with that popularity is the issue of effective management and exploration of CCTV cameras to ensure security and safety for people. With a huge data recorded from cameras, we need systems that support monitoring and analysis of camera data as well as instant video data retrieval. In this paper, we introduce a system to contextualize the video content recorded from CCTV cameras in real-time and store data efficiently. The big data generated from the system will make it easier to monitor and operate the camera system to access and search for objects and content in the surveillance cameras.

This paper proposes an antenna performance prediction model in the autonomous driving radar manufacturing process. Our research work is based upon a challenge dataset, Driving Radar Manufacturing Process Dataset, and a typical AutoML machine learning workflow engine, Pycaret open-source Python library. Note that the dataset contains the total 70 data-items, out of which 54 used as input features and 16 used as output features, and the dataset is properly built into resolving the multi-output regression problem. During the data regression analysis and preprocessing phase, we identified several input features having similar correlations and so detached some of those input features, which may become a serious cause of the multicollinearity problem that affect the overall model performance. In the training phase, we train each of output-feature regression models by using the AutoML approach. Next, we selected the top 5 models showing the higher performances in the AutoML result reports and applied the ensemble method so as for the selected models' performances to be improved. In performing the experimental performance evaluation of the regression prediction model, we particularly used two metrics, MAE and RMSE, and the results of which were 0.6928 and 1.2065, respectively. Additionally, we carried out a series of experiments to verify the proposed model's performance by comparing with other existing models' performances. In conclusion, we enhance accuracy for safer autonomous vehicles, reduces manufacturing costs through AutoML-Pycaret and machine learning ensembled model, and prevents the production of faulty radar systems, conserving resources. Ultimately, the proposed model holds significant promise not only for antenna performance but also for improving manufacturing quality and advancing radar systems in autonomous vehicles.

In the field of process mining, the discovery of process patterns from event logs remains a challenging topic and has always interested many researchers. Exploiting the process model remains a major challenge and is highly dependent on event log characteristics, such as dataset size, the completeness of the event trace, and especially the complexity of the process model structure. The $\rho $ (rho)-algorithm is a powerful process mining algorithm that can mine all the structured information control net (SICN), such as sequential, selective, parallel, and iterative. The $\rho $ -algorithm also faces challenges when mining complex process patterns, consisting of many SICN primitive patterns combined at multiple levels that satisfy properties matched pairing and the proper nesting. This paper presents a new approach, defines the formalization of the multi-level and compound control-flow gateways (ML-CCFG) and a series of rules for decision-making these gateways, and proposes an algorithm extending from $\rho $ -algorithm (we named it $\rho $ MC-algorithm), that can efficiently discover the SICN-oriented process model containing ML-CCFG from process instances event log. We developed an implementation system $\rho $ MC-algorithm to discover and visualize in a graphical form SICN-oriented process models from the datasets of the IEEE XES-formatted process enactment event logs. We also perform a series of experimental analyses using the implemented system on the process enactment event log datasets to verify the proposed algorithm.

Efficient allocating human resource obtains significant advantages to business operations in enterprises. Manipulating the available human resources of a business optimizes the expense of developing products and services, improves work productivity, and brings more value to the business. In the process of creating products and benefits in the enterprise, a network of working relationships is formed among employees in the organization. In this paper, we propose an approach to support decision-making on human resource allocation in an organization based on the human-centric network discovered from the event logs. To be more precise, the knowledge derived from human-centric networks formed through the operation phases of the information systems is used to make resource allocation strategies in the organizations. Using the appropriateness probabilities and social network metrics for allocating human resources, we clarify efficiently allocating resources based on different metrics. The system architecture and the scenarios are also described for planning human resources allocation based on these networks. Furthermore, we use real-life data set for getting experimental verification and proving the proposal. This approach will be a promising solution to support businesses effectively allocating resources in their organizations.

The purpose of this paper is to verify the functional correctness of a specific process mining algorithm and to validate the requirement satisfaction of the targeted algorithm, as well. The functional requirement of the process mining algorithm is to discover all the structural process patterns, such as linear (sequential), disjunctive (selective-OR), conjunctive (parallel-AND), and repetitive (iterative-LOOP) process patterns, from a dataset of process enactment event logs, and to eventually build a structured business process model by assembling all the discovered structural process patterns. The targeted algorithm to be verified and validated in this paper is called as ρ-Algorithm that is especially devised for discovering a structured information control net process model (SICN-oriented process model) from a dataset especially prepared in the standardized IEEE XES event stream format. In order to carry out the verification and validation, we have successfully developed a process mining system based upon the ρ-Algorithm, and through which four rounds of experiments are taken with four real datasets that are well fitted into the following situational mining and discovering scenarios: Faultlessness Process, Matched-Pairing Violation but Sound Process, Matched-Pairing and Proper-Nesting Violation but Sound Process, and Nebulous Process (Adaptive Cases). Conclusively, we strongly believe that the ρ-Algorithm is theoretically safe as well as functionally operable, and that the implemented ρ-Algorithm is also practically applicable to a process mining system as one of the business process and workflow intelligence solutions.