Search

Different electrical and thermoelectric properties of a Si-based thermoelectric generator (TEG) are described based on the Kubo–Greenwood formalism. Temperature and doping dependence, phonon scattering (acoustic and optical phonons), and scattering on impurities are included. Comparisons with experimentally verified data confirm the validity of the model. Experimental studies were carried out on a micromechanically fabricated TEG. Devices were realized using a standard CMOS SOI technology in a lateral geometry. All thermopiles are located on a thin membrane to reduce the heat flow. The thickness of the membrane was adjusted between 20 and 30 µm ensuring also sufficient mechanical stability. Measurements on individual devices confirm the results of the theoretical model. The Seebeck coefficient was calculated and experimentally measured as S = 0.5 mV/K at an acceptor level of 1019 cm−3 at room temperature. The power factor is S2 · σ = 0.0073 W/mK2.

Abstract The challenge of sustaining user engagement in eHealth interventions is a pressing issue with significant implications for the effectiveness of these digital health tools. This study investigates user engagement in a cognitive-behavioral therapy-based eHealth intervention for procrastination, using a dataset from a randomized controlled trial of 233 university students. Various machine learning models, including Decision Tree, Gradient Boosting, Logistic Regression, Random Forest, and Support Vector Machines, were employed to predict patterns of user engagement. The study adopted a two-phase analytical approach. In the first phase, all features of the dataset were included, revealing ‘total_minutes’—the total time participants spent on the intervention and the eHealth platform—as the most significant predictor of engagement. This finding emphasizes the intuitive notion that early time spent on the platform and the intervention is a strong indicator of later user engagement. However, to gain a deeper understanding of engagement beyond this predominant metric, the second phase of the analysis excluded ‘total_minutes’. This approach allowed for the exploration of the roles and interdependencies of other engagement indicators, such as ‘number_intervention_answersheets’—the number of completed lessons, ‘logins_first_4_weeks’—login frequency, and ‘number_diary_answersheets’—the number of completed diaries. The results from this phase highlighted the multifaceted nature of engagement, showing that while ‘total_minutes’ is strongly correlated with engagement, indicating that more engaged participants tend to spend more time on the intervention, the comprehensive engagement profile also depends on additional aspects like lesson completions and frequency of platform interactions.

This review investigates persuasive design frameworks within eHealth, concentrating on methodologies, their prevalence in mental and behavioral health applications, and identifying current research gaps. An extensive search was conducted across 8 databases, focusing on English publications with full text available. The search prioritized primary research articles, post-2011 applications, and eHealth platforms emphasizing treatment or support. The inclusion process was iterative, involving multiple authors, and relied on detailed criteria to ensure the relevance and contemporaneity of selected works. The final review set comprised 161 articles, providing an overview of persuasive design frameworks in eHealth. The review highlights the state of the art in the domain, emphasizing the utilization and effectiveness of these frameworks in eHealth platforms. This review details the restricted adoption of persuasive design frameworks within the field of eHealth, particularly in the mental and behavioral sectors. Predominant gaps include the scarcity of comparative evaluations, the underrepresentation of tailored interventions, and the unclear influence of persuasive components on user experience. There is a notable requirement for further scrutiny and refinement of persuasive design frameworks. Addressing these concerns promises a more substantial foundation for persuasive design in eHealth, potentially enhancing user commitment and platform efficiency.

Abstract Since the beginning of the COVID-19 pandemic, many different machine learning models have been developed to detect and verify COVID-19 pneumonia based on chest X-ray images. Although promising, binary models have only limited implications for medical treatment, whereas the prediction of disease severity suggests more suitable and specific treatment options. In this study, we publish severity scores for the 2358 COVID-19 positive images in the COVIDx8B dataset, creating one of the largest collections of publicly available COVID-19 severity data. Furthermore, we train and evaluate deep learning models on the newly created dataset to provide a first benchmark for the severity classification task. One of the main challenges of this dataset is the skewed class distribution, resulting in undesirable model performance for the most severe cases. We therefore propose and examine different augmentation strategies, specifically targeting majority and minority classes. Our augmentation strategies show significant improvements in precision and recall values for the rare and most severe cases. While the models might not yet fulfill medical requirements, they serve as an appropriate starting point for further research with the proposed dataset to optimize clinical resource allocation and treatment.

Abstract In medical imaging, deep learning models can be a critical tool to shorten time-to-diagnosis and support specialized medical staff in clinical decision making. The successful training of deep learning models usually requires large amounts of quality data, which are often not available in many medical imaging tasks. In this work we train a deep learning model on university hospital chest X-ray data, containing 1082 images. The data was reviewed, differentiated into 4 causes for pneumonia, and annotated by an expert radiologist. To successfully train a model on this small amount of complex image data, we propose a special knowledge distillation process, which we call Human Knowledge Distillation. This process enables deep learning models to utilize annotated regions in the images during the training process. This form of guidance by a human expert improves model convergence and performance. We evaluate the proposed process on our study data for multiple types of models, all of which show improved results. The best model of this study, called PneuKnowNet, shows an improvement of + 2.3% points in overall accuracy compared to a baseline model and also leads to more meaningful decision regions. Utilizing this implicit data quality-quantity trade-off can be a promising approach for many scarce data domains beyond medical imaging.

Abstract Mobile applications have gained popularity in healthcare in recent years. These applications are an increasingly important pillar of public health care, as they open up new possibilities for data collection and can lead to new insights into various diseases and disorders thanks to modern data analysis approaches. In this context, Ecological Momentary Assessment (EMA) is a commonly used research method that aims to assess phenomena with a focus on ecological validity and to help both the user and the researcher observe these phenomena over time. One phenomenon that benefits from this capability is the chronic condition tinnitus. TrackYourTinnitus (TYT) is an EMA-based mobile crowdsensing platform designed to provide more insight into tinnitus by repeatedly assessing various dimensions of tinnitus, including perception (i.e., perceived presence). Because the presence of tinnitus is the dimension that is of great importance to chronic tinnitus patients and changes over time in many tinnitus patients, we seek to predict the presence of tinnitus based on the not directly related dimensions of mood, stress level, arousal, and concentration level that are captured in TYT. In this work, we analyzed a dataset of 45,935 responses to a harmonized EMA questionnaire using different machine learning techniques. In addition, we considered five different subgroups after consultation with clinicians to further validate our results. Finally, we were able to predict the presence of tinnitus with an accuracy of up to 78% and an AUC of up to 85.7%.

Integrating the Internet of Things (IoT) into business process management (BPM) aims to increase the automation level, efficiency, transparency, and comprehensibility of the business processes taking place in the physical world. The IoT enables the seamless networking of physical devices, allowing for the enrichment of processes with real-time data about the physical world and, thus, for optimized process automation and monitoring. To realize these benefits, the modeling of IoT-aware processes needs to be appropriately supported. Despite the great attention paid to this topic, more clarity is needed about the current state of the art of corresponding modeling solutions. Capturing IoT characteristics in business process models visually or based on labels is essential to ensure effective design and communication of IoT-aware business processes. A clear discernibility of IoT characteristics can enable the precise modeling and analysis of IoT-aware processes and facilitate collaboration among different stakeholders. With an increasing number of process model elements, it becomes crucial that process model readers can understand the IoT aspects of business processes in order to make informed decisions and to optimize the processes with respect to IoT integration. This paper presents the results of a large user study (N = 249) that explored the perception of IoT aspects in BPMN 2.0 process models to gain insights into the IoT’s involvement in business processes that drive the successful implementation and communication of IoT-aware processes.

Notifications are an essential part of the user experience on smart mobile devices. While some apps have to notify users immediately after an event occurs, others can schedule notifications strategically to notify them only on opportune moments. This tailoring allows apps to shorten the users’ interaction delay. In this paper, we present the results of a comprehensive study that identified the factors that influence users’ interaction delay to their smartphone notifications. We analyzed almost 10 million notifications collected in-the-wild from 922 users and computed their response times with regard to their demographics, their Big Five personality trait scores and the device’s charging state. Depending on the app category, the following tendencies can be identified over the course of the day: Most notifications were logged in late morning and late afternoon. This number decreases in the evening, between 8 p.m. and 11 p.m., and at the same time exhibits the lowest average interaction delays at daytime. We also found that the user’s sex and age is significantly associated with the response time. Based on the results of our study, we encourage developers to incorporate more information on the user and the executing device in their notification strategy to notify users more effectively.

The study of complex process models often encounters challenges in terms of comprehensibility. This paper explores using modularization as a strategy to mitigate such challenges, notably the reduction in complexity. Previous research has delved into the comprehensibility of modularized process models, yet an unresolved question about the cognitive factors at play during their comprehension still needs to be answered. Addressing the latter, the paper presents findings from an innovative study combining eye-tracking and concurrent think-aloud techniques involving 25 participants. The study aimed to comprehend how individuals comprehend process models when presented in three different modular formats: flattened process models, models with grouped elements, and models with subprocesses. The results shed light on varying comprehension strategies employed by participants when navigating through these modularized process models. The paper concludes by suggesting avenues for future research guided by these insights.

Data scientists in the Natural Language Processing (NLP) field confront the challenge of reconciling the necessity for data-centric analyses with the imperative to safeguard sensitive information, all while managing the substantial costs linked to the collection process of training data. In a Federated Learning (FL) system, these challenges can be alleviated by the training of a global model, eliminating the need to centralize sensitive data of clients. However, distributed NLP data is usually Non-Independent and Identically Distributed (Non-IID), which leads to poorer generalizability of the global model when trained with Federated Averaging (FedAvg). Recently proposed extensions to FedAvg promise to improve the global model performance on Non-IID data. Yet, such advanced FL algorithms trained on multilingual Non-IID texts have not been studied in industry and academia in detail. This paper compares, for the first time, the FL algorithms: FedAvg, FedAvgM, FedYogi, FedAdam and FedAdagrad for a binary text classification task using 12078 tailored real-world news reports in English, Portuguese, Spanish and Hindi. For this objective, pre-trained DistilBERT and BERT models fine-tuned with these texts are used. The paper results show that FedYogi is the most stable and robust FL algorithm when DistilBERT is used, achieving an average macro F1 score of 0.7789 for IID and 0.7755 for Non-IID protest news. The study also exhibits that BERT models trained with weighted FedAvg and FedAvgM can achieve a similar prediction power as centralized language models, demonstrating the potential of leveraging FL in the NLP domain without the need to collect data centrally.

As mobile devices have become a central part of our daily lives, they are also becoming increasingly important in research. In the medical context, for example, smartphones are used to collect ecologically valid and longitudinal data using Ecological Momentary Assessment (EMA), which is mostly implemented through questionnaires delivered via smart notifications. This type of data collection is intended to capture a patient’s condition on a moment-to-moment and longer-term basis. To collect more objective and contextual data and to understand patients even better, researchers can not only use patients’ input via EMA, but also use sensors as part of the Mobile Crowdsensing (MCS) approach. In this paper, we examine how researchers have embraced the topic of MCS in the context of EMA through a systematic literature review. This PRISMA-guided review is based on the databases PubMed, Web of Science, and EBSCOhost. It is shown through the results that both EMA research in general and the use of sensors in EMA research are steadily increasing. In addition, most of the studies reviewed used mobile apps to deliver EMA to participants, used a fixed-time prompting strategy, and used signal-contingent or interval-contingent self-assessment as sampling/assessment strategies. The most commonly used sensors in EMA studies are the accelerometer and GPS. In most studies, these sensors are used for simple data collection, but sensor data are also commonly used to verify study participant responses and, less commonly, to trigger EMA prompts. Security and privacy aspects are addressed in only a subset of mHealth EMA publications. Moreover, we found that EMA adherence was negatively correlated with the total number of prompts and was higher in studies using a microinteraction-based EMA (μEMA) approach as well as in studies utilizing sensors. Overall, we envision that the potential of the technological capabilities of smartphones and sensors could be better exploited in future, more automated approaches.

In medical imaging, deep learning models serve as invaluable tools for expediting diagnoses and aiding specialized medical professionals in making clinical decisions. However, effectively training deep learning models typically necessitates substantial quantities of high-quality data, a resource often lacking in numerous medical imaging scenarios. One way to overcome this deficiency is to artificially generate such images. Therefore, in this comparative study we train five generative models to artificially increase the amount of available data in such a scenario. This synthetic data approach is evaluated on a a downstream classification task, predicting four causes for pneumonia as well as healthy cases on 1082 chest X-ray images. Quantitative and medical assessments show that a Generative Adversarial Network (GAN)-based approach significantly outperforms more recent diffusion-based approaches on this limited dataset with better image quality and pathological plausibility. We show that better image quality surprisingly does not translate to improved classification performance by evaluating five different classification models and varying the amount of additional training data. Class-specific metrics like precision, recall, and F1-score show a substantial improvement by using synthetic images, emphasizing the data rebalancing effect of less frequent classes. However, overall performance does not improve for most models and configurations, except for a DreamBooth approach which shows a +0.52 improvement in overall accuracy. The large variance of performance impact in this study suggests a careful consideration of utilizing generative models for limited data scenarios, especially with an unexpected negative correlation between image quality and downstream classification improvement.

The aim of this study was to investigate the impact of different coping styles on situational coping in everyday life situations and gender differences. An ecological momentary assessment study with the mobile health app TrackYourStress was conducted with 113 participants. The coping styles Positive Thinking, Active Stress Coping, Social Support, Support in Faith, and Alcohol and Cigarette Consumption of the Stress and Coping Inventory were measured at baseline. Situational coping was assessed by the question “How well can you cope with your momentary stress level” over 4 weeks. Multilevel models were conducted to test the effects of the coping styles on situational coping. Additionally, gender differences were evaluated. Positive Thinking (p = 0.03) and Active Stress Coping (p = 0.04) had significant positive impacts on situational coping in the total sample. For women, Social Support had a significant positive effect on situational coping (p = 0.046). For men, Active Stress Coping had a significant positive effect on situational coping (p = 0.001). Women had higher scores on the SCI scale Social Support than men (p = 0.007). These results suggest that different coping styles could be more effective in daily life for women than for men. Taking this into account, interventions tailored to users’ coping styles might lead to better coping outcomes than generalized interventions.

Industrial data scarcity is one of the largest factors holding back the widespread use of machine learning in manufacturing. To overcome this problem, the concept of transfer learning was developed and has received much attention in recent industrial research. This paper focuses on the problem of time series segmentation and presents the first in-depth research on transfer learning for deep learning-based time series segmentation on the industrial use case of end-of-line pump testing. In particular, we investigate whether the performance of deep learning models can be increased by pretraining the network with data from other domains. Three different scenarios are analyzed: source and target data being closely related, source and target data being distantly related, and source and target data being non-related. The results demonstrate that transfer learning can enhance the performance of time series segmentation models with respect to accuracy and training speed. The benefit can be most clearly seen in scenarios where source and training data are closely related and the number of target training data samples is lowest. However, in the scenario of non-related datasets, cases of negative transfer learning were observed as well. Thus, the research emphasizes the potential, but also the challenges, of industrial transfer learning.

The Internet of Things (IoT) enables a variety of smart applications, including smart home, smart manufacturing, and smart city. By enhancing Business Process Management Systems with IoT capabilities, the execution and monitoring of business processes can be significantly improved. Providing a holistic support for modeling, executing and monitoring IoT-driven processes, however, constitutes a challenge. Existing process modeling and process execution languages, such as BPMN 2.0, are unable to fully meet the IoT characteristics (e.g., asynchronicity and parallelism) of IoT-driven processes. In this article, we present BPMNE4IoT—A holistic framework for modeling, executing and monitoring IoT-driven processes. We introduce various artifacts and events based on the BPMN 2.0 metamodel that allow realizing the desired IoT awareness of business processes. The framework is evaluated along two real-world scenarios from two different domains. Moreover, we present a user study for comparing BPMNE4IoT and BPMN 2.0. In particular, this study has confirmed that the BPMNE4IoT framework facilitates the support of IoT-driven processes.

Predictive process monitoring (PPM) has been discussed as a use case of process mining for several years. PPM enables foreseeing the future of an ongoing business process by predicting, for example, relevant information on the way in which running processes terminate or on related process performance indicators. A large share of PPM approaches adopt Machine Learning (ML), taking advantage of the accuracy and precision of ML models. Consequently, PPM inherits the challenges of traditional ML approaches. One of these challenges concerns the need to gain user trust in the generated predictions. This issue is addressed by explainable artificial intelligence (XAI). However, in addition to ML characteristics, the choices made and the techniques applied in the context of PPM influence the resulting explanations. This necessitates the availability of a study on the effects of different choices made in the context of a PPM task on the explainability of the generated predictions. In order to address this gap, we systemically investigate the effects of different PPM settings on the data fed into an ML model and subsequently into the employed XAI method. We study how differences between the resulting explanations indicate several issues in the underlying data. Example of these issues include collinearity and high dimensionality of the input data. We construct a framework for performing a series of experiments to examine different choices of PPM dimensions (i.e., event logs, preprocessing configurations, and ML models), integrating XAI as a fundamental component. In addition to agreements, the experiments highlight several inconsistencies between data characteristics and important predictors used by the ML model on one hand, and explanations of predictions of the investigated ML model on the other.

Predictive Process Monitoring (PPM) has been integrated into process mining use cases as a value-adding task. PPM provides useful predictions on the future of the running business processes with respect to different perspectives, such as the upcoming activities to be executed next, the final execution outcome, and performance indicators. In the context of PPM, Machine Learning (ML) techniques are widely employed. In order to gain trust of stakeholders regarding the reliability of PPM predictions, eXplainable Artificial Intelligence (XAI) methods have been increasingly used to compensate for the lack of transparency of most of predictive models. Multiple XAI methods exist providing explanations for almost all types of ML models. However, for the same data, as well as, under the same preprocessing settings or same ML models, generated explanations often vary significantly. Corresponding variations might jeopardize the consistency and robustness of the explanations and, subsequently, the utility of the corresponding model and pipeline settings. This paper introduces a framework that enables the analysis of the impact PPM-related settings and ML-model-related choices may have on the characteristics and expressiveness of the generated explanations. Our framework provides a means to examine explanations generated either for the whole reasoning process of an ML model, or for the predictions made on the future of a certain business process instance. Using well-defined experiments with different settings, we uncover how choices made through a PPM workflow affect and can be reflected through explanations. This framework further provides the means to compare how different characteristics of explainability methods can shape the resulting explanations and reflect on the underlying model reasoning process.

The ubiquity of mobile devices fosters the combined use of ecological momentary assessments (EMA) and mobile crowdsensing (MCS) in the field of healthcare. This combination not only allows researchers to collect ecologically valid data, but also to use smartphone sensors to capture the context in which these data are collected. The TrackYourTinnitus (TYT) platform uses EMA to track users’ individual subjective tinnitus perception and MCS to capture an objective environmental sound level while the EMA questionnaire is filled in. However, the sound level data cannot be used directly among the different smartphones used by TYT users, since uncalibrated raw values are stored. This work describes an approach towards making these values comparable. In the described setting, the evaluation of sensor measurements from different smartphone users becomes increasingly prevalent. Therefore, the shown approach can be also considered as a more general solution as it not only shows how it helped to interpret TYT sound level data, but may also stimulate other researchers, especially those who need to interpret sensor data in a similar setting. Altogether, the approach will show that measuring sound levels with mobile devices is possible in healthcare scenarios, but there are many challenges to ensuring that the measured values are interpretable.