The recent emergence of novel hardware platforms, such as quantum computers and Digital/CMOS annealers, capable of solving combinatorial optimization problems has spurred interest in formulating key problems as Ising models, a mathematical abstraction shared by a number of these platforms. In this work, we focus on constrained clustering, a semi-supervised learning task that involves using limited amounts of labelled data, formulated as constraints, to improve clustering accuracy. We present an Ising modeling framework that is flexible enough to support various types of constraints and we instantiate the framework with two common types of constraints: pairwise instance-level and partition-level. We study the proposed framework, both theoretically and empirically, and demonstrate how constrained clustering problems can be solved on a specialized CMOS annealer. Empirical evaluation across eight benchmark sets shows that our framework outperforms the state-of-theart heuristic algorithms and that, unlike those algorithms, it can solve problems that involve combinations of constraint types. We also show that our framework provides high quality solutions orders of magnitudes more quickly than a recent constraint programming approach, making it suitable for mainstream data mining tasks.

Discovery plays a key role in data-driven analysis of business processes. The vast majority of contemporary discovery algorithms aims at the identification of control-flow constructs. The increase in data richness, however, enables discovery that incorporates the context of process execution beyond the control-flow perspective. A "control-flow first" approach, where context data serves for refinement and annotation, is limited and fails to detect fundamental changes in the control-flow that depend on context data. In this work, we thus propose a novel approach for combining the control-flow and data perspectives under a single roof by extending inductive process discovery. Our approach provides criteria under which context data, handled through unsupervised learning, take priority over control-flow in guiding process discovery. The resulting model is a process tree, in which some operators carry data semantics instead of control-flow semantics. We evaluate the approach using synthetic and real-world datasets and show that the resulting models are superior to state-of-the-art discovery methods in terms of measures that are based on multi perspective alignments.

A significant challenge in declarative approaches to scheduling is the creation of a model: the set of resources and their capacities and the types of activities and their temporal and resource requirements. In practice, such models are developed manually by skilled consultants and used repeatedly to solve different problem instances. For example, in a factory, the model may be used each day to schedule the current customer orders. In this work, we aim to automate the creation of such models by learning them from event data. We introduce a novel methodology that combines process mining, timed Petri nets (TPNs), and constraint programming (CP). The approach learns a sub-class of TPN from event logs of executions of past schedules and maps the TPN to a broad class of scheduling problems. We show how any problem of the scheduling class can be converted to a CP model. With new instance data (e.g., the day’s orders), the CP model can then be solved by an off-the-shelf solver. Our approach provides an end-to-end solution, going from event logs to model-based optimal schedules. To demonstrate the value of the methodology we conduct experiments in which we learn and solve scheduling models from two types of data: logs generated from job-shop scheduling benchmarks and real-world event logs from an outpatient hospital.

Time prediction is an essential component of decision making in various Artificial Intelligence application areas, including transportation systems, healthcare, and manufacturing. Predictions are required for efficient resource allocation and scheduling, optimized routing, and temporal action planning. In this work, we focus on time prediction in congested systems, where entities share scarce resources. To achieve accurate and explainable time prediction in this setting, features describing system congestion (e.g., workload and resource availability), must be considered. These features are typically gathered using process knowledge, (i.e., insights on the interplay of a system’s entities). Such knowledge is expensive to gather and may be completely unavailable. In order to automatically extract such features from data without prior process knowledge, we propose the model of congestion graphs, which are grounded in queueing theory. We show how congestion graphs are mined from raw event data using queueing theory based assumptions on the information contained in these logs. We evaluate our approach on two real-world datasets from healthcare systems where scarce resources prevail: an emergency department and an outpatient cancer clinic. Our experimental results show that using automatic generation of congestion features, we get an up to 23% improvement in terms of relative error in time prediction, compared to common baseline methods. We also detail how congestion graphs can be used to explain delays in the system.

Information systems record the execution of transactions as part of business processes in event logs. Process mining analyses such event logs, e.g., by discovering process models. Recently, various discovery algorithms have been proposed, each with specific advantages and limitations. In this work, we argue that, instead of relying on a single algorithm, the outcomes of different algorithms shall be fused to combine the strengths of individual approaches. We propose a general framework for such fusion and instantiate it with two new discovery algorithms: The Exhaustive Noise-aware Inductive Miner (exNoise), which, exhaustively searches for model improvements; and the Adaptive Noise-aware Inductive Miner (adaNoise), a computationally tractable version of exNoise. For both algorithms, we formally show that they outperform each of the individual mining algorithms used by them. Our empirical evaluation further illustrates that fusion-based discovery yields models of better quality than state-of-the-art approaches.

With the increasing availability of business process related event logs, the scalability of techniques that discover a process model from such logs becomes a performance bottleneck. In particular, exploratory analysis that investigates manifold parameter settings of discovery algorithms, potentially using a software-as-a-service tool, relies on fast response times. However, common approaches for process model discovery always parse and analyse all available event data, whereas a small fraction of a log could have already led to a high-quality model. In this paper, we therefore present a framework for process discovery that relies on statistical pre-processing of an event log and significantly reduce its size by means of sampling. It thereby reduces the runtime and memory footprint of process discovery algorithms, while providing guarantees on the introduced sampling error. Experiments with two public real-world event logs reveal that our approach speeds up state-of-the-art discovery algorithms by a factor of up to 20 .

Analysing performance of business processes is an important vehicle to improve their operation. Specifically, an accurate assessment of sojourn times and remaining times enables bottleneck analysis and resource planning. Recently, methods to create respective performance models from event logs have been proposed. These works are severely limited, though: They either consider control-flow and performance information separately, or rely on an ad-hoc selection of temporal relations between events. In this paper, we introduce the Temporal Network Representation (TNR) of a log, based on Allen’s interval algebra, as a complete temporal representation of a log, which enables simultaneous discovery of control-flow and performance information. We demonstrate the usefulness of the TNR for detecting (unrecorded) delays and for probabilistic mining of variants when modelling the performance of a process. In order to compare different models from the performance perspective, we develop a framework for measuring performance fitness. Under this framework, we provide guarantees that TNR-based process discovery dominates existing techniques in measuring performance characteristics of a process. To illustrate the practical value of the TNR, we evaluate the approach against three real-life datasets. Our experiments show that the TNR yields an improvement in performance fitness over state-of-the-art algorithms.

Predictive process monitoring is concerned with predicting measures of interest for a running case (e.g., a business outcome or the remaining time) based on historical event logs. Most of the current predictive process monitoring approaches only consider intra-case information that comes from the case whose measures of interest one wishes to predict. However, in many systems, the outcome of a running case depends on the interplay of all cases that are being executed concurrently. For example, in many situations, running cases compete over scarce resources. In this paper, following standard predictive process monitoring approaches, we employ supervised machine learning for prediction. In particular, we present a method for feature encoding of process cases that relies on a bi-dimensional state space representation: the first dimension corresponds to intra-case dependencies, while the second dimension reflects inter-case dependencies to represent shared information among running cases. The inter-case encoding derives features based on the notion of case types that can be used to partition the event log into clusters of cases that share common characteristics. To demonstrate the usefulness and applicability of the method, we evaluated it against two real-life datasets coming from an Israeli emergency department process, and an open dataset of a manufacturing process.

Operational process models such as generalised stochastic Petri nets (GSPNs) are useful when answering performance queries on business processes (e.g. ‘how long will it take for a case to finish?’). Recently, methods for process mining have been developed to discover and enrich operational models based on a log of recorded executions of processes, which enables evidence-based process analysis. To avoid a bias due to infrequent execution paths, discovery algorithms strive for a balance between over-fitting and under-fitting regarding the originating log. However, state-of-the-art discovery algorithms address this balance solely for the control-flow dimension, neglecting possible over-fitting in terms of performance annotations. In this work, we thus offer a technique for performance-driven model reduction of GSPNs, using structural simplification rules. Each rule induces an error in performance estimates with respect to the original model. However, we show that this error is bounded and that the reduction in model parameters incurred by the simplification rules increases the accuracy of process performance prediction. We further show how to find an optimal sequence of applying simplification rules to obtain a minimal model under a given error budget for the performance estimates. We evaluate the approach with a real-world case in the healthcare domain, showing that model simplification indeed yields significant improvements in time prediction accuracy.

While models and event logs are readily available in modern organizations, their quality can seldom be trusted. Raw event recordings are often noisy, incomplete, and contain erroneous recordings. The quality of process models, both conceptual and data-driven, heavily depends on the inputs and parameters that shape these models, such as domain expertise of the modelers and the quality of execution data. The mentioned quality issues are specifically a challenge for conformance checking. Conformance checking is the process mining task that aims at coping with low model or log quality by comparing the model against the corresponding log, or vice versa. The prevalent assumption in the literature is that at least one of the two can be fully trusted. In this work, we propose a generalized conformance checking framework that caters for the common case, when one does neither fully trust the log nor the model. In our experiments we show that our proposed framework balances the trust in model and log as a generalization of state-of-the-art conformance checking techniques.

Process mining is a rapidly developing field that aims at automated modeling of business processes based on data coming from event logs. In recent years, advances in tracking technologies, e.g., Real-Time Locating Systems (RTLS), put forward the ability to log business process events as location sensor data. To apply process mining techniques to such sensor data, one needs to overcome an abstraction gap, because location data recordings do not relate to the process directly. In this work, we solve the problem of mapping sensor data to event logs based on process knowledge. Specifically, we propose interactions as an intermediate knowledge layer between the sensor data and the event log. We solve the mapping problem via optimal matching between interactions and process instances. An empirical evaluation of our approach shows its feasibility and provides insights into the relation between ambiguities and deviations from process knowledge, and accuracy of the resulting event log.

The performance of scheduled business processes is of central importance for services and manufacturing systems. However, current techniques for performance analysis do not take both queueing semantics and the process perspective into account. In this work, we address this gap by developing a novel method for utilizing rich process logs to analyze performance of scheduled processes. The proposed method combines simulation, queueing analytics, and statistical methods. At the heart of our approach is the discovery of an individual-case model from data, based on an extension of the Colored Petri Nets formalism. The resulting model can be simulated to answer performance queries, yet it is computational inefficient. To reduce the computational cost, the discovered model is projected into Queueing Networks, a formalism that enables efficient performance analytics. The projection is facilitated by a sequence of folding operations that alter the structure and dynamics of the Petri Net model. We evaluate the approach with a real-world dataset from Dana-Farber Cancer Institute, a large outpatient cancer hospital in the United States.

Service processes, for example in transportation, telecommunications or the health sector, are the backbone of today’s economies. Conceptual models of such service processes enable operational analysis that supports, e.g., resource provisioning or delay prediction. Automatic mining of such operational models becomes feasible in the presence of event-data traces. In this work, we target the mining of models that assume a resource-driven perspective and focus on queueing effects. We propose a solution for the discovery and validation problem of scheduled service processes - processes with a predefined schedule for the execution of activities. Our prime example for such processes are complex outpatient treatments that follow prior appointments. Given a process schedule and data recorded during process execution, we show how to discover Fork/Join networks, a specific class of queueing networks, and how to assess their operational validity. We evaluate our approach with a real-world dataset comprising clinical pathways of outpatient clinics, recorded by a real-time location system (RTLS). We demonstrate the value of the approach by identifying and explaining operational bottlenecks.

In service processes, as found in the telecommunications, financial, or healthcare sector, customers compete for the scarce capacity of service providers. For such processes, performance analysis is important and it often targets the time that customers are delayed prior to service. However, this wait time cannot be fully explained by the load imposed on service providers. Indeed, it also depends on resource scheduling protocols, which determine the order of activities that a service provider decides to follow when serving customers. This work focuses on automatically learning resource decisions from events. We hypothesize that queueing information serves as an essential element in mining such protocols and hence, we utilize the queueing perspective of customers in the mining process. We propose two types of mining techniques: advanced classification methods from data mining that include queueing information in their explanatory features and heuristics that originate in queueing theory. Empirical evaluation shows that incorporating the queueing perspective into mining of scheduling protocols improves predictive power.

Information systems have been widely adopted to support service processes in various domains, e.g., in the telecommunication, finance, and health sectors. Information recorded by systems during the operation of these processes provides an angle for operational process analysis, commonly referred to as process mining. In this work, we establish a queueing perspective in process mining to address the online delay prediction problem, which refers to the time that the execution of an activity for a running instance of a service process is delayed due to queueing effects. We present predictors that treat queues as first-class citizens and either enhance existing regression-based techniques for process mining or are directly grounded in queueing theory. In particular, our predictors target multi-class service processes, in which requests are classified by a type that influences their processing. Further, we introduce queue mining techniques that derive the predictors from event logs recorded by an information system during process execution. Our evaluation based on large real-world datasets, from the telecommunications and financial sectors, shows that our techniques yield accurate online predictions of case delay and drastically improve over predictors neglecting the queueing perspective.