Residual inhibition (RI) is a phenomenon observed in many tinnitus patients, where tinnitus remains temporarily suppressed for a short duration—typically less than a minute-after the cessation of an appropriate masking stimulus. Despite decades of clinical interest in RI, machine learning (ML)-based, feature-driven classification approaches remain scarce. In this study, we investigate the potential of ML models to classify RI by developing a dedicated data analysis pipeline. Given the heterogeneous nature of the features—including numerical, binary, and ordinal variables-we apply feature importance techniques tailored for mixed-type data to ensure a comprehensive evaluation and improve interpretability. Our results demonstrate a clear separation between RI classes, highlighting the relevance of specific clinical and audiological factors in distinguishing them. Building on this, we assess the predictive power of RI classifications with high confidence within a supervised learning framework to determine their relevance for treatment outcome prediction. While our findings confirm that RI can be effectively classified, they also suggest that RI alone is not sufficient to serve as a reliable predictor for treatment outcomes.

The respiratory system can be significantly affected by thoracic injuries, which can lead to complications such as lung dysfunction. Therefore, immediate diagnosis, along with the precise location of these injuries is crucial, as it allows targeted medical interventions, reduces unnecessary treatments, and accelerates patient recovery. Respiratory function is dependent on the diaphragm and intercostal muscles, which work in sync to produce individual breathing motion. These motions are highly individual and can be influenced by injuries and respiratory therapy. In this paper, we employ an embedded sensor network to record human breathing patterns and present a novel approach utilizing Temporal Graph Neural Networks (TGCNs) to develop radiation-free, non-invasive techniques for breathing motion monitoring. We modeled the sensor network into a graph structure in which the nodes represent the sensors and the edges represent the correlation among them, enabling our method to effectively capture both spatial and temporal relationships between sensor measurements. We simulate a network of standard sensors to monitor human breathing movements, generating a synthetic dataset of different breathing motions using real data for oversampling, including the degree of abnormalities such as mild, moderate, and severe. We perform node-level classification using the Graph Convolutional Gated Recurrent Unit (GConvGRU) model to identify the abnormal breathing pattern along with the location and severity level of the injuries. Our results demonstrate that TGCNs can accurately localize breathing abnormalities through a graphical sensor network representation, facilitating the location of the potential severity of the injury, and improving remote diagnosis, particularly in post-injury rehabilitation.

BACKGROUND: Current clinical decision support systems (DSS) are trained and validated on observational data from the clinic in which the DSS is going to be applied. This is problematic for treatments that have already been validated in a randomized clinical trial (RCT), but have not yet been introduced in any clinic. In this work, we report on a method for training and validating the DSS core before introduction to a clinic, using the RCT data themselves. The key challenges we address are of missingness, foremost: missing rationale when assigning a treatment to a patient (the assignment is at random), and missing verification evidence, since the effectiveness of a treatment for a patient can only be verified (ground truth) if the treatment was indeed assigned to the patient - but then the assignment was at random. MATERIALS: We use the data of a multi-armed clinical trial that investigated the effectiveness of single treatments and combination treatments for 240+ tinnitus patients recruited and treated in 5 clinical centres. METHODS: To deal with the 'missing rationale for treatment assignment' challenge, we re-model the target variable that measures the outcome of interest, in order to suppress the effect of the individual treatment, which was at random, and control on the effect of treatment in general. To deal with missing features for many patients, we use a learning core that is robust to missing features. Further, we build ensembles that parsimoniously exploit the small patient numbers we have for learning. To deal with the 'missing verification evidence' challenge, we introduce counterfactual treatment verification, a verification scheme that juxtaposes the effectiveness of the recommendations of our approach to the effectiveness of the RCT assignments in the cases of agreement/disagreement between the two. RESULTS AND LIMITATIONS: We demonstrate that our approach leverages the RCT data for learning and verification, by showing that the DSS suggests treatments that improve the outcome. The results are limited through the small number of patients per treatment; while our ensemble is designed to mitigate this effect, the predictive performance of the methods is affected by the smallness of the data. OUTLOOK: We provide a basis for the establishment of decision supporting routines on treatments that have been tested in RCTs but have not yet been deployed clinically. Practitioners can use our approach to train and validate a DSS on new treatments by simply using the RCT data available to them. More work is needed to strengthen the robustness of the predictors. Since there are no further data available to this purpose, but those already used, the potential of synthetic data generation seems an appropriate alternative.

Health apps have the potential to collect data in real life, which makes them a useful tool. Users can monitor themselves, and researchers can gain comprehensive insights. However, user behavior is subject to fluctuations. On the one hand, this makes it complicated to use machine learning techniques, and on the other hand, there is a need to predict activity to stimulate it when necessary.This paper investigates user participation, focusing on their usage behavior, possible patterns of engagement, and resulting neighborhoods. The goal is to use these insights to predict a user’s engagement. Three ways of modeling patterns of engagement are presented. Neighborhood methods are used to infer conclusions about future behavior from the information of other users. Three methods are proposed to predict the length of a phase of inactivity, the length of the next phase of activity, and its values and structure.Applied to two real-world datasets, the results show that the approaches are suitable for making these predictions.These findings can be used to understand better and predict the behavior of mHealth users and plan possible interventions. This research expands the boundaries of research using self-monitoring mHealth apps, as user behavior can now be more easily formalized and applied. Furthermore, the approach can help to influence user engagement in a targeted way. Thus, this contribution exceeds our previous state-of-the-art work in its possibilities.

Learning is an essential skill that humans and animals need to interact successfully with stimuli in their surrounding environment. Discovering the learning motifs applied by humans and animals is a highly complex task. Indeed, identifying significant motifs in behavioural actions by analyzing how actions are affected by stimuli and prior actions enables us to understand how a learning motif is formed by a subject. Granger causality (GC) is a statistical tool used to check whether the past of a variable X is predictive of the future of another variable Y, in this case, we say that X Granger causes Y. Furthermore, subjects may also change the learning motif followed over time due to the learning process. In this study, we propose a method that (1) models a ‘learning motif’ as a set of Granger causality relationships involving past stimuli and past actions, (2) employs a hidden Markov model (HMM) to capture the change in the followed learning motifs and (3) identifies the salient learning motifs that are common to many subjects. The evaluation of our proposed method is not a trivial task due to the absence of the ground truth of the learning motifs. In general, it is difficult to acquire the ground truth of the learning motifs because for example, animals cannot articulate these learning motifs and this also applies to humans in complex learning tasks. Therefore, we also propose a solution that does not require ground truth to validate the derived learning motifs. We evaluate our proposed method on behavioural data collected from two groups of mice, a group of healthy mice and a group of mice with induced cognitive impairment. We show that our model is appropriate for identifying the learning motifs followed by these two groups of animals during a learning experiment in which animals are expected to maximize the reward gain.

Smartphones and other mobile devices offer a valu-able opportunity to gather patient-specific health data during everyday life. However, the increasing popularity of mobile health apps demands specialized data analysis methods that can handle the unique, patient-based, time-dependent, and often multivariate data collected by these apps. This work explores the analysis of patient-based mHealth data to develop personalized prediction models. The models can incor-porate data not only from the individual patient, but also from other similar patients using patient-specific neighborhoods. Our approach entails selecting the appropriate data for a particular prediction task and dataset. We also discuss when to utilize data from other patients and offer guidance on selecting similarity functions, models, and model combinations. The approach is illustrated on the case study of tinnitus, a perception of sound without an external source, which can be highly distressing. Its presentation and treatment success are patient-specific. As part of the UNITI project, daily diary data of the patients is collected. Evaluation favored the use of personalized models using patient-specific neighborhoods over a global model using all data, or only using a patient's own data for tinnitus distress prediction.

Abstract In clinical practice, algorithmic predictions may seriously jeopardise patients’ health and thus are required to be validated by medical experts before a final clinical decision is met. Towards that aim, there is need to incorporate explainable artificial intelligence techniques into medical research. In the specific field of epileptic seizure detection there are several machine learning algorithms but less methods on explaining them in an interpretable way. Therefore, we introduce XAI4EEG: an application-aware approach for an explainable and hybrid deep learning-based detection of seizures in multivariate EEG time series. In XAI4EEG, we combine deep learning models and domain knowledge on seizure detection, namely (a) frequency bands, (b) location of EEG leads and (c) temporal characteristics. XAI4EEG encompasses EEG data preparation, two deep learning models and our proposed explanation module visualizing feature contributions that are obtained by two SHAP explainers, each explaining the predictions of one of the two models. The resulting visual explanations provide an intuitive identification of decision-relevant regions in the spectral, spatial and temporal EEG dimensions. To evaluate XAI4EEG, we conducted a user study, where users were asked to assess the outputs of XAI4EEG, while working under time constraints, in order to emulate the fact that clinical diagnosis is done - more often than not - under time pressure. We found that the visualizations of our explanation module (1) lead to a substantially lower time for validating the predictions and (2) leverage an increase in interpretability, trust and confidence compared to selected SHAP feature contribution plots.

An accurate assessment of the cardiovascular system and prediction of cardiovascular diseases (CVDs) are crucial. Cardiac blood flow data provide insights about patient-specific hemodynamics. However, there is a lack of machine learning approaches for a feature-based classification of heart-healthy people and patients with CVDs. In this paper, we investigate the potential of morphological and hemodynamic features extracted from measured blood flow data in the aorta to classify heart-healthy volunteers (HHV) and patients with bicuspid aortic valve (BAV). Furthermore, we determine features that distinguish male vs. female patients and elderly HHV vs. BAV patients. We propose a data analysis pipeline for cardiac status classification, encompassing feature selection, model training, and hyperparameter tuning. Our results suggest substantial differences in flow features of the aorta between HHV and BAV patients. The excellent performance of the classifiers separating between elderly HHV and BAV patients indicates that aging is not associated with pathological morphology and hemodynamics. Our models represent a first step towards automated diagnosis of CVS using interpretable machine learning models.

Applications in mobile health (mHealth) empower self-monitoring of chronic conditions of the user and also offer insights to medical experts. The data generated by these apps constitute one time series per user. These time series vary substantially in length and contain ‘gaps’, as users pause or stop interacting with the app. In order to design measures that promote patient engagement with the app, it is necessary to predict and understand decline in engagement. We measured the performance of the algorithms on two real-world datasets from an mHealth app. We show that all approaches outperform the baseline and that shapelet, dictionary and matrix distance approach perform similarly for long-term prediction. This is particularly important because it allows early intervention towards increase of engagement. In this paper, we present an approach that uses the missingness information to process time series with large gaps.

Missing features can negatively impact the performance of machine learning solutions and past research has been focused on how to acquire the most predictive features in static scenarios. Active Feature Acquisition (AFA) for data streams extends the conventional static paradigm by taking into account that the importance of a feature may change as the stream drifts.In this study, we propose an AFA method that takes the cost of the features for each arriving instance into account and, at the same time, allows multiple features to be acquired at once. This reflects the fact that labels often depend on multiple features, so that acquiring many low-cost features may result in higher improvement than acquiring a single feature, as has been proposed in our earlier work [1]. We evaluated our approach on 7 real and 8 synthetic data sets. We investigated three different budget sizes and three different feature cost configurations for 7 different percentages of missing data on each of our data sets. We use the results of our experiments to elaborate on when the acquisition of multiple features is more or less beneficial than acquiring a single feature. Finally, we show the need for more sophisticated metrics to estimate a feature’s predictive quality.

Adherence and phases of non-adherence in the usage of mHealth apps for self-monitoring generate time series characterized by gaps of varying duration. These reduce the knowledge that can be gained from the data because important observations are not captured. In this paper, an approach is presented that allows experts to estimate the possible duration of a user’s absence in order to build on it and take appropriate action.For this purpose, the users’ time series (with gaps and of different lengths) are decomposed into sequences that have no gaps anymore. Each sequence is associated with the duration of the subsequent gap. Using unsupervised binning, these gap values are sorted into a predefined number of categories. Sequences with similar gap durations are thus assigned the same category labels, grouped using unsupervised time series clustering and are assigned with cluster labels. Thus triplets are derived consisting of user ID, cluster label and gap category label. These triplets can then be used for collaborative filtering with matrix factorization. It is now possible to estimate the gap duration even if the same sequence has not yet been observed for the app user.The results show that the quality of binning depends on the appropriate choice of the number of categories, on the technique used, and on the maximum length of the gaps. In this example, a 5-star rating, the Fischer-Jenks algorithm and a maximum gap length of 30 days. We demonstrate how clustering can be used to gradually adapt the time series to the conditions of collaborative filtering and how complex matrix factorization models respond better to the complex structures of the data and lead to better results. In this work, K-Medoids and Agglomerative Clustering obtained applicable results and for matrix factorization SVD++.We believe that our approach provides a valuable back-end tool for experts to better assess the adherence of users to a selfmonitoring app.

Ecological momentary assessment (EMA) has been used in many mHealth apps. EMA captures valuable insights into many diseases. Identifying the Granger causal relationships across the EMA variables may contribute to the interpretation of a disease and improve treatment decisions. In our study, we perform a circadian conditional Granger causality analysis on multivariate time series. The analysis was done on EMA data of 270 users of an mHealth app on tinnitus and on their registration data, using the latter to explain the circadian conditional Granger causal relationships. We discovered that some EMA items Granger cause others for more than 8% of the mHealth app users and that these users have answered 8 out of 40 questions at registration differently.

A stream of users' interactions with an mHealth app can be seen as the result of a stochastic process that can be captured by an algorithm that learns over the whole stream. But is it only one process? We investigate to what extend learning for each user separately delivers better predictions than learning one model over the whole stream. Our application scenario is the prediction of Ecological Momentary Assessments (EMA) for an mHealth app (TinnitusTipps) on tinnitus. The data were recorded as part of a pilot study, in which one group of users received non-personalized suggestions (tips) throughout the study, while the other group received tips only during the second half of the study. Our method encompasses user-centric and global stream learning for EMA prediction, combined under a Contextual Multi-Armed Bandit (CMAB) that captures the context of each user group and incorporates the prediction quality of each learner into the reward function. We show that user-centric learning is beneficial for users who contribute many EMA, while a learner over the whole stream is better for users with few EMA.

Understanding the behaviour of online communities on social media platforms helps in controlling the accompanying adverse phenomena like false claims and conspiracy theories. Towards that aim, we present a multifaceted analysis of Twitter users and set a special focus on their 5G coronavirus tweets, in the light of the infamous 5G conspiracy theory. We contrast their idiosyncrasies with general coronavirus tweets in five million tweets acquired over a period of four months and a cadence of two weeks. We find that awareness raising and situation monitoring are two main concerns of Twitter online community. Additionally, 0.13% of the elicited tweets mention 5G, where a causal relationship between 5G and coronavirus besides China-related discussions are more evident compared to overall tweets. Indicators of other conspiracy theories appear more frequently in 5G coronavirus tweets as well.

Mobile Health (mhealth) applications are increasing in popularity, and the collection of disease-specific time series data using Ecological Momentary Assessment (EMA) questionnaires has been shown to help in the creation of personalised predictors for next-step forecasting, which can be crucial in giving preemptive interventions. In this work, we propose a framework that aims to mitigate a common issue in EMA data - that some users contribute a bulk of the data while most users contribute too little. Our proposed framework aims to discover a `useful' neighbourhood of `long' users for `short' ones, by optimising for the error of the user-level predictors for users with little data available for learning. For each user-level predictor, this is done by iteratively adding the next-most-similar long user from a similarity-ordered list as long as the error of the learned model does not increase. This method is compared against a baseline that exploits all available data for the long users, as well as an exhaustive search model that retains only only those users that yield the lowest error predictor. We also explore multiple ways to define similarity, and study the impact of each on the two search strategies on two EMA datasets from an mHealth app `TrackYourDiabetes' - with users from Bulgaria and Spain. Our experiments over the two datasets show a 2.5% and 36.6% improvement respectively for RMSE while using on average 42.8% and 69.9% less data than the baseline method.

Electronic health records (EHR) often include multiple perspectives on a patient's current state of well-being (e.g. vital signs and subjective indicators measured by questionnaires). In this study, we use these perspectives to build phenotypes of chronic tinnitus patients and investigate how these phenotypes are associated with response to treatment. Therefore, we model patients as nodes in a network, where those perspectives are interpreted as layers of a multi-layer network. To identify phenotypes of patients in the network, we implement a community detection algorithm. Some of these communities can be considered as phenotypes if they represent subgroups of patients that are similar according to the investigated perspectives. Furthermore, we analyze the influence of the layers on the final community structure of patients. We then propose a method to add layers given their community structure similarity. Finally, we fit a model, per community, to predict the treatment outcome. In some communities, this prediction outperformed the baseline scenario where the predictor was fitted to all patients.

Abstract Some mHealth apps record user activity continuously and unobtrusively, while other apps rely by nature on user engagement and self-discipline: users are asked to enter data that cannot be assessed otherwise, e.g., on how they feel and what non-measurable symptoms they have. Over time, this leads to substantial differences in the length of the time series of recordings for the different users. In this study, we propose two algorithms for wellbeing-prediction from such time series, and we compare their performance on the users of a pilot study on diabetic patients - with time series length varying between 8 and 87 recordings. Our first approach learns a model from the few users, on which many recordings are available, and applies this model to predict the 2nd, 3rd, and so forth recording of users newly joining the mHealth platform. Our second approach rather exploits the similarity among the first few recordings of newly arriving users. Our results for the first approach indicate that the target variable for users who use the app for long are not predictive for users who use the app only for a short time. Our results for the second approach indicate that few initial recordings suffice to inform the predictive model and improve performance considerably.

Missing data is a common occurrence in the time series domain, for instance due to faulty sensors, server downtime or patients not attending their scheduled appointments. One of the best methods to impute these missing values is Multiple Imputations by Chained Equations (MICE) which has the drawback that it can only model linear relationships among the variables in a multivariate time series. The advancement of deep learning and its ability to model non-linear relationships among variables make it a promising candidate for time series imputation. This work proposes a modified Convolutional Denoising Autoencoder (CDA) based approach to impute multivariate time series data in combination with a preprocessing step that encodes time series data into 2D images using Gramian Angular Summation Field (GASF). We compare our approach against a standard feed-forward Multi Layer Perceptron (MLP) and MICE. All our experiments were performed on 5 UEA MTSC multivariate time series datasets, where 20 to 50% of the data was simulated to be missing completely at random. The CDA model outperforms all the other models in 4 out of 5 datasets and is tied for the best algorithm in the remaining case.

Abstract Attrition or dropout is the most severe missingness problem in longitudinal cohort study data where some participants do not show up for follow‐up examinations. Dropouts result in biased data and cause the reduction of 1ata set size. Moreover, they limit the power of statistical analysis and the validity of study findings. Visualization can play a strong role in analysing and displaying the missingness patterns. In this work, we present VIVID, a framework for the visual analysis of missing values in cohort study data. VIVID is inspired by discussions with epidemiologists and adds visual components to their current statistics‐based approaches. VIVID provides functions for exploration, imputation and validity check of imputations. The main focus of this paper is multiple imputation to fix the missing data.

Smart devices and low-powered sensors are becoming increasingly ubiquitous and nowadays almost all of these devices are connected, which is a promising foundation for crowdsensing of data related to various environmental phenomena. Resulting data is especially meaningful when it is related to time and location. Interestingly, many existing approaches built their solution on monolithic backends that process data on a per-request basis. However, for many scenarios, such technical setting is not suitable for managing data requests of a large crowd. For example, when dealing with millions of data points, still many challenges arise for modern smartphones if calculations or advanced visualization features must be accomplished directly on the smartphone. Therefore, the work at hand proposes an architectural design for managing geospatial data of tinnitus patients, which combines a cloudnative approach with Big Data concepts used in the Internet of Things. The presented architectural design shall serve as a generic foundation to implement (1) a scalable backend for a platform that covers the aforementioned crowdsensing requirements as well as to provide (2) a sophisticated stream processing concept to calculate and pre-aggregate incoming measurement data of tinnitus patients. Following this, this paper presents a visualization feature to provide users with a comprehensive overview of noise levels in their environment based on noise measurements. This shall help tinnitus or hearing-impaired patients to avoid locations with a burdensome sound level.

What are the basic forms of healthcare data? How are Electronic Health Records and Cohorts structured? How can we identify the key variables in such data and how important are temporal abstractions? What are the main challenges in knowledge extraction from medical data sources? What are the key machine algorithms used for this purpose? What are the main questions that clinicians and medical experts pose to machine learning researchers?

Treatment of patients with tinnitus is mainly pursued on the basis of screening data, encompassing answers to questionnaires or audiological examinations. Since tinnitus affects the quality of life of patients and is associated with comorbidities like depression, the screening step involves the acquisition of extensive amounts of information, which may contribute to the design of a personalized treatment. Often times, it can be observed that patients give up their treatment before completion (i.e., they constitute “dropout” cases). In this study, we investigate how multi-target classification on the screening data can contribute to characterize patients that will drop out of the study. For our analysis, we base or considerations on the target variable “tinnitus loudness”, i.e. the subjectively perceived loudness of the phantom signal. Following this, we attempt to identify variables that explain the tinnitus loudness together with the likelihood of interrupting the treatment. To be more precise, we report on results from gathered data of 1419 tinnitus patients from the University Hospital of Regensburg.

Epidemiological population-based studies collect hundreds of socio-demographic, lifestyle-related, and health related variables for thousands of individuals in order to better characterize health and disease in a defined population. To understand the relations between the variables in the study data, we employ Bayesian Networks, as they not only represent associations between variables but also assign probabilities to these associations. The probabilistic associations allow us to draw inference for unknown events in question, based on the provided evidence. In our work, we induce a Bayesian Network from the data of the population-based epidemiological Study of Health in Pomerania (SHIP), to identify variables related to the outcome "fatty liver". We report on Bayesian Network structure learning, identification of variables associated with the outcome, and the strong associations identified among the variables.

Mobile applications can help patients with a chronical disease to record their Ecological Momentary Assessments (EMA) and to get a more precise impression of how their disease manifests itself during day and night and over longer time periods. Such crowdsensing applications contribute to patient empowerment, in which patients monitor their disease and, sometimes, learn to cope better with it. An open question is whether physicians can also be helped in assisting their patients, by understanding similarities and differences in the patients' evolution. We study the EMA of patients with the chronical disease tinnitus, as recorded with the mobile crowdsensing application Track Your Tinnitus. We propose a method that captures similarities in patient evolution, taking account of the differences in the frequency of each patient's EMA recordings. We incorporate this method into a complete workflow that encompasses following components: an algorithm that captures similarities among patients on the basis of their registration data, a method that juxtaposes static patient similarity to EMA-based patient similarity, and a method that identifies those subspaces of the static feature space and those of the EMA-based feature space, which are mainly contributing to patient similarity. We report on our results for the time period recordings from 2014 till 2017 of 450 tinnitus patients from TrackYourTinnitus mobile application.

Intracranial aneurysms are pathologic dilations of the vessel wall, which bear the risk of rupture and of fatal consequences for the patient. Since treatment may be accompanied by severe complications as well, rupture risk assessment and thus rupture risk prediction plays an important role in clinical research. In this work, we investigate the potential of morphological features for rupture risk status classification in 100 intracranial aneurysms. We propose a pipeline for morphological feature extraction and rupture status classification with subsequent feature ranking and inspection. Our classification setup involves training separate models for each aneurysm type (sidewall or bifurcation) with multiple learning algorithms. We report on the classification performance of our pipeline and examine the predictive power of each morphological parameter towards rupture status classification. Further, we identify the most important features for the best models and study their marginal prediction.

Presently, mHealth technology is often applied in the context of chronic diseases to gather data that may lead to new and valuable medical insights. As many aspects of chronic diseases are not completely understood, new data sources might be promising. mHealth technology may help in this context as it can be easily used in everyday life. Moreover, the bring your own device principle encourages many patients to use their smartphone to learn more about their disease. The less is known about a disorder (e.g., tinnitus), the more patients crave for new insights and opportunities. Despite the fact that existing mHealth technology like mobile crowdsensing has already gathered data that may help patients, in general, less is known whether and how data gathered with different mobile technologies may differ. In this context, one relevant aspect is the contribution of the mobile operating system itself. For example, are there differences between Android and iOS users that utilize the same mHealth technology for a disease. In the TrackYourTinnitus project, a mobile crowdsensing mHealth platform was developed to gather data for tinnitus patients in order to reveal new insights on this disorder with high economic and patient-related burdens. As many data sets were gathered during the last years that enable us to compare Android and iOS users, the work at hand compares characteristics of these users. Interesting insights like the one that Android users with tinnitus are significantly older than iOS users could be revealed by our study. However, more evaluations are necessary for TrackYourTinnitus in particular and mHealth technology in general to understand how smartphones affect the gathering of data on chronic diseases when using them in the large.

Diabetic foot syndrome is a frequent and serious complication occurring among patients with diabetes. In this study, we investigate the potential of intelligent wearables that monitor temperature changes of the foot surface through temperature sensors. In particular, we are interested in identifying differences between the temperature variations recorded on patients with the disorder and healthy people during an experiment. To this purpose, we propose a method that encompasses shapelet-based timeseries classification and shapelet ranking on predictiveness. We report on our results for an experiment consisting of stance and rest periods of increasing duration.

Huge amounts of textual streams are generated nowadays, especially in social networks like Twitter and Facebook. As the discussion topics and user opinions on those topics change drastically with time, those streams undergo changes in data distribution, leading to changes in the concept to be learned, a phenomenon called concept drift. One particular type of drift, that has not yet attracted a lot of attention is feature drift, i.e., changes in the features that are relevant for the learning task at hand. In this work, we propose an approach for handling feature drifts in textual streams. Our approach integrates i) an ensemble-based mechanism to accurately predict the feature/word values for the next time-point by taking into account the different features might be subject to different temporal trends and ii) a sketch-based feature space maintenance mechanism that allows for a memory-bounded maintenance of the feature space over the stream. Experiments with textual streams from the sentiment analysis, email preference and spam detection demonstrate that our approach achieves significantly better or competitive performance compared to baselines.

Stream classification algorithms traditionally treat arriving observations as independent. However, in many applications the arriving examples may depend on the "entity" that generated them, e.g. in product reviewing or in the interactions of users with an application server. In this study, we investigate the potential of this dependency by partitioning the original stream of observations into entity-centric substreams and by incorporating entity-specific information into the learning model. We propose a k Nearest Neighbour inspired stream classification approach (kNN), in which the label of an arriving observation is predicted by exploiting knowledge on the observations belonging to this entity and to entities similar to it. For the computation of entity similarity, we consider knowledge about the observations and knowledge about the entity, potentially transferred from another domain. To distinguish between cases where this kind of knowledge transfer is beneficial for stream classification and cases where the knowledge on the entities does not contribute to classifying the observations, we also propose a heuristic approach based on random sampling of substreams using k Random Entities (kRE). Our learning scenario is not fully supervised: after acquiring labels for the initial few observations of each entity, we assume that no additional labels arrive, and attempt to predict the labels of near-future and far-future observations from that initial seed. We report on our findings from three datasets.

Crowdsourcing is a popular mechanism used for labeling tasks to produce large corpora for training. However, producing a reliable crowd labeled training corpus is challenging and resource consuming. Research on crowdsourcing has shown that label quality is much affected by worker engagement and expertise. In this study, we postulate that label quality can also be affected by inherent ambiguity of the documents to be labeled. Such ambiguities are not known in advance, of course, but, once encountered by the workers, they lead to disagreement in the labeling - a disagreement that cannot be resolved by employing more workers. To deal with this problem, we propose a crowd labeling framework: we train a disagreement predictor on a small seed of documents, and then use this predictor to decide which documents of the complete corpus should be labeled and which should be checked for document-inherent ambiguities before assigning (and potentially wasting) worker effort on them. We report on the findings of the experiments we conducted on crowdsourcing a Twitter corpus for sentiment classification.

Subgroup discovery (SD) exploits its full value in applications where the goal is to generate understandable models. Epidemiologists search for statistically significant relationships between risk factors and outcome in large and heterogeneous datasets encompassing information about the participants health status gathered from questionnaires, medical examinations and image acquisition. SD algorithms can help epidemiologists by automatically detecting such relationships presented as comprehensible rules, aiming to ultimately improve prevention, diagnosis and treatment of diseases. However, SD algorithms often produce large and overlapping rule sets requiring the expert to conduct a manual post-filtering step that is time-consuming and tedious. In this work, we propose a clustering-based algorithm that hierarchically reorganizes rule sets and summarizes all important concepts while maintaining diversity between the rule clusters. For each cluster, a representative rule is selected and then displayed to the expert who in turn can drill-down to other cluster members. We evaluate our algorithm on two cohort study datasets where the diseases hepatic steatosis and goiter serve as target variable, respectively. We report on our findings with respect to effectiveness of our algorithm and we present selected subpopulations.

Many symptoms of neuropsychiatric disorders such as tinnitus are subjective and vary over time. Usually, in interviews or self-report questionnaires, patients are asked to report symptoms as well as their severity and duration retrospectively. However, only little is known to what degree such retrospective reports reflect the symptoms experienced in daily life some time ago. Mobile technologies can help to bridge this gap: mobile self-help services allow patients to record their symptoms prospectively when (or shortly after) they occur in daily life. In this study, we present results that we obtained with the mobile crowdsensing platform TrackYourTinnitus to show that there is a discrepancy between the prospective assessment of symptom variability and the retrospective report thereof. To be more precise, we evaluated the real-time entries provided to the platform by individuals experiencing tinnitus. The results indicate that mobile technologies like the TrackYourTinnitus crowdsensing platform may go beyond the role of an assistive service for patients by contributing to more accurate diagnosis and, hence, to a more elaborated treatment.

Personalized medicine benefits from the identification of subpopulations that exhibit higher prevalence of a disease than the general population: such subpopulations can become the target of more intensive investigations to identify risk factors and to develop dedicated therapies. Classification rule discovery algorithms are an appropriate tool for discovering such subpopulations: they scale well, even for multi-dimensional data and deliver comprehensible patterns. However, they may generate hundreds of rules and thus call for exploration methods. In this study, we extend the tool Interactive Medical Miner for the discovery of classification rules, into the Interactive Classification rule Explorer ICE, which offers functionalities for rule exploration, grouping, rule visualization and statistics. We report on our first results for the classification of cohort data on goiter, a disorder of the thyroid gland.

The identification of subpopulations with particular characteristics with respect to a disease is important for personalized diagnostics and therapy design. For some diseases, the outcome is described by more than one target variable. An example is tinnitus: the perceived loudness of the phantom signal and the level of distress caused by it are both relevant targets for diagnosis and therapy. In this work, we study the potential of multi-target classification for the identification of those screening variables, which separate best among the different subpopulations of patients, paying particular attention to subpopulations with discordant value combinations of loudness and distress. We analyse the screening data of 1344 tinnitus patients from the University Hospital Regensburg, including questions from 7 questionnaires, and report on the performance of our workflow in target separation and in ranking the questionnaires variables on their discriminative power.

The identification of predictive features associated with distinct medical outcomes is a key requirement for meaningful clinical decision support. Usually, their discovery is based on sets of labeled examples and an analysis of the inherent information of the features w. r. t. the target variable. However, obtaining large sets of labeled examples may be not feasible and the sole label consideration could even dilute characteristics unique to distinct subgroups. In such cases, instead of considering the value of the target variable, expert knowledge on the similarity between examples could be utilized. In this work we propose a new algorithm for the "Discovery of Relevant Example-constrained Subspaces" (DRESS) which uses constraints on the similarity between examples to discover feature sets that describe a target concept. DRESS exploits the density of clusters and the distance-behavior between constrained examples to evaluate the quality of a feature set without requiring explicit information about the target variable. We evaluate DRESS against classical feature selection methods on cohort participants for the disorder "hepatic steatosis", and report on our findings on classifier performance and identified important features.

The diabetic foot syndrome (DFS) is linked to loss of neuron functions, implying that the patients do not feel their feet and may unknowingly injure themselves or apply excessive plantar pressure. Such patients are at 17-40 times higher risk of foot amputation than non-diabetics. Sensor-equipped insoles are being developed to warn diabetics against inadverted excessive pressure. For the successful use of such technology, it is essential to understand how patients distribute plantar pressure load and to identify common pressure patterns, to be later used as basis for recognizing abnormalities. In this study, we propose a mining workflow for the discovery of pressure patterns among DFS patients. Our approach encompasses different ways of modeling pressure distribution among foot regions, and workpaths for the computation of similarity between patients and the construction of clusters of patients who apply pressure on their feet the same way. We report on our findings from a dataset of experiment participants who wore sensor-equipped insoles and were asked to apply and release pressure repeatedly over a time period of several minutes. We elaborate on the pressure patterns thus identified and juxtapose them to findings from the literature.

This work addresses active learning for multi-class classification. Active learning algorithms optimize classifier performance by successively selecting the most beneficial instances from a pool of unlabeled instances to be labeled by an oracle. In this work, we study the influence of the following factors for active learning: (1) an instance's impact, (2) its posterior, and (3) the reliability of this posterior. To do so, we propose a new decision-theoretic approach, called multi-class probabilistic active learning (McPAL). Building on a probabilistic active learning framework, our approach is non-myopic, fast, and optimizes a performance measure (like accuracy) directly. Considering all influence factors, McPAL determines the expected gain in performance to compare the usefulness of instances. For this purpose, it calculates the density weighted expectation over the true posterior and over all possible labeling combinations in a closed-form solution. Thus, in contrast to other multi-class algorithms, it considers the posterior's reliability which improved the performance. In our experimental evaluation, we show that the combination of the selected influence factors works best and that McPAL is superior in comparison to various other multi-class active learning algorithms on six datasets.

Today’s competitive marketplace requires the industry to understand unique and particular needs of their customers. Product line practices enable companies to create individual products for every customer by providing an interdependent set of features. Users configure personalized products by consecutively selecting desired features based on their individual needs. However, as most features are interdependent, users must understand the impact of their gradual selections in order to make valid decisions. Thus, especially when dealing with large feature models, specialized assistance is needed to guide the users in configuring their product. Recently, recommender systems have proved to be an appropriate mean to assist users in finding information and making decisions. In this paper, we propose an advanced feature recommender system that provides personalized recommendations to users. In detail, we offer four main contributions: (i) We provide a recommender system that suggests relevant features to ease the decision-making process. (ii) Based on this system, we provide visual support to users that guides them through the decision-making process and allows them to focus on valid and relevant parts of the configuration space. (iii) We provide an interactive open-source configurator tool encompassing all those features. (iv) In order to demonstrate the performance of our approach, we compare three different recommender algorithms in two real case studies derived from business experience.

Medical research can greatly benefit from advances in data mining. We propose a mining approach for cohort analysis in a longitudinal population-based epidemiological study, and show that modelling and exploiting the evolution of cohort participants over time improves classification quality towards an outcome (a disease). Our mining workflow encompasses steps for tracing the evolution of the cohort participants and for using evolution features in classification. We show that our approach separates better between classes and that change in the values of variables is predictive. We report on results for the liver disorder hepatic steatosis (high fat accumulation in the liver), but our approach is appropriate for classification of longitudinal epidemiological data on further disorders.

In year 2015, we experience a proliferation of scientific publications, conferences and funding programs on KDD for medicine and healthcare. However, medical scholars and practitioners work differently from KDD researchers: their research is mostly hypothesis-driven, not data-driven. KDD researchers need to understand how medical researchers and practitioners work, what questions they have and what methods they use, and how mining methods can fit into their research frame and their everyday business. Purpose of this tutorial is to contribute to this learning process. We address medicine and healthcare; there the expertise of KDD scholars is needed and familiarity with medical research basics is a prerequisite. We aim to provide basics for (1) mining in epidemiology and (2) mining in the hospital. We also address, to a lesser extent, the subject of (3) preparing and annotating Electronic Health Records for mining.

Clinical decision support relies on the findings of epidemiological (longitudinal and cross-sectional) studies on predictive features and risk factors for diseases. Such features flow into the diagnostic procedures. Personalized medicine, which aims to optimize clinical decision making by taking individual characteristics of the patients into account, relies on the findings of epidemiology on groups of cohort participants that have common risk factors and exhibit the outcome under study. The identification of such groups requires modeling and exploiting similarity among individuals described through medical tests. In this work, we study how similarity measures for complex objects contribute to class separation for a multifactorial disorder. We present a data preparation, partitioning and classification workflow on cohort participants for the disorder "hepatic steatosis", and report on our findings on classifier performance and identified important features.

Active Mining of Big Data requires fast approaches that ideally select for a user-specified performance measure and arbitrary classifier the optimal instance for improving the classification performance. Existing generic approaches are either slow, like error reduction, or heuristics, like uncertainty sampling. We propose a novel, fast yet versatile approach that directly optimises any user-specified performance measure: Probabilistic Active Learning (PAL).

We investigate the task of breast tumor classification based on dynamic contrast-enhanced magnetic resonance image data (DCE-MRI). Our objective is to study how the formation of regions of similar voxels contributes to distinguishing between benign and malignant tumors. First, we perform clustering on each tumor with different algorithms and parameter settings, and then combine the clustering results to identify the most suspect region of the tumor and derive features from it. With these features we train classifiers on a set of tumors that are difficult to classify, even for human experts. We show that the features of the most suspect region alone cannot distinguish between benign and malignant tumors, yet the properties of this region are indicative of tumor malignancy for the dataset we studied.

Scholars in life sciences have to process huge amounts of data in a disciplined and efficient way. These data are spread among thousands of databases which overlap in content but differ substantially with respect to interface, formats and data structure. Search engines have the potential of assisting in data retrieval from these structured sources but fall short of providing a relevance ranking of the results that reflects the needs of life science scholars. One such need is to acquire insights to cross-references among entities in the databases, whereby search hits with many cross-references are expected to be more informative than those with few cross-references. In this work, we investigate to what extend this expectation holds. We propose BioXREF, a method that extracts cross-references from multiple life science databases by combining targeted crawling, pointer chasing, sampling and information extraction. We study the retrieval quality of our method and the relationship between manually crafted relevance ranking and relevance ranking based on cross-references, and report on first, promising results.

Getting an overview of a large online social net-work and deciding which communities to join is a challenging task for a new user. We propose a method that maps a large network into a smaller graph with two kinds of nodes: a node of the first kind is representative of a community, a node of the second kind is neighbor to a representative and rejects the semantics of that community. Our approach encompasses a learning and ranking algorithm that derives this smaller graph from the original one, and a visualization algorithm that returns a graph layout to the observer. We report on our results on inspecting the network of a folksonomy.

We introduce the concern of confidentiality protection of business information for the publication of search engine query logs and derived data. We study business confidentiality, as the protection of nonpublic data from institutions, such as companies and people in the public eye. In particular, we relate this concern to the involuntary exposure of confidential Web site information, and we transfer this problem into the field of privacy-preserving data mining. We characterize the possible adversaries interested in disclosing Web site confidential data and the attack strategies that they could use. These attacks are based on different vulnerabilities found in query log for which we present several anonymization heuristics to prevent them. We perform an experimental evaluation to estimate the remaining utility of the log after the application of our anonymization techniques. Our experimental results show that a query log can be anonymized against these specific attacks while retaining a significant volume of useful data.

Document collections evolve over time, new topics emerge and old ones decline. At the same time, the terminology evolves as well. Much literature is devoted to topic evolution in finite document sequences assuming a fixed vocabulary. In this study, we propose “Topic Monitor” for the monitoring and understanding of topic and vocabulary evolution over an infinite document sequence, i.e. a stream. We use Probabilistic Latent Semantic Analysis (PLSA) for topic modeling and propose new folding-in techniques for topic adaptation under an evolving vocabulary. We extract a series of models, on which we detect index-based topic threads as human-interpretable descriptions of topic evolution.

Detecting densely connected subgroups in graphs such as communities in social networks is of interest in many research fields. Several methods have been developed to find communities but most of them have a high time complexity and are thus not applicable for large networks. Inspired by the clustering algorithm incremental DBSCAN we propose a density-based graph clustering algorithm DENGRAPH that is designed to deal with large dynamic datasets with noise and present first experimental results.

There is much recent work on detecting and tracking change in clusters, often based on the study of the spatiotemporal properties of a cluster. For the many applications where cluster change is relevant, among them customer relationship management, fraud detection and marketing, it is also necessary to provide insights about the nature of cluster change: Is a cluster corresponding to a group of customers simply disappearing or are its members migrating to other clusters? Is a new emerging cluster reflecting a new target group of customers or does it rather consist of existing customers whose preferences shift? To answer such questions, we propose the framework MONIC for modeling and tracking of cluster transitions. Our cluster transition model encompasses changes that involve more than one cluster, thus allowing for insights on cluster change in the whole clustering. Our transition tracking mechanism is not based on the topological properties of clusters, which are only available for some types of clustering, but on the contents of the underlying data stream. We present our first results on monitoring cluster transitions over the ACM digital library.

A social network consists of people who interact in some way such as members of online communities sharing information via the WWW. To learn more about how to facilitate community building e.g. in organizations, it is important to analyze the interaction behavior of their members over time. So far, many tools have been provided that allow for the analysis of static networks and some for the temporal analysis of networks - however only on the vertex and edge level. In this paper we propose two approaches to analyze the evolution of two different types of online communities on the level of subgroups. The first method consists of statistical analyses and visualizations that allow for an interactive analysis of subgroup evolutions in communities that exhibit a rather membership structure. The second method is designed for the detection of communities in an environment with highly fluctuating members. For both methods, we discuss results of experiments with real data from an online student community

Web-usage mining has become the subject of intensive research, as its potential for personalized services, adaptive Web sites and customer profiling is recognized. However, the reliability of Web-usage mining results depends heavily on the proper preparation of the input datasets. In particular, errors in the reconstruction of sessions and incomplete tracing of users’ activities in a site can easily result in invalid patterns and wrong conclusions. In this study, we evaluate the performance of heuristics employed to reconstruct sessions from the server log data. Such heuristics are called to partition activities first by user and then by visit of the user in the site, where user identification mechanisms, such as cookies, may or may not be available. We propose a set of performance measures that are sensitive to two types of reconstruction errors and appropriate for different applications in knowledge discovery (KDD) applications. We have tested our framework on the Web server data of a frame-based Web site. The first experiment concerned a specific KDD application and has shown the sensitivity of the heuristics to particularities of the site's structure and traffic. The second experiment is not bound to a specific application but rather compares the performance of the heuristics for different measures and thus for different application types. Our results show that there is no single best heuristic, but our measures help the analyst in the selection of the heuristic best suited for the application at hand.

With the increase in the size of data sets, data mining has recently become an important research topic and is receiving substantial interest from both academia and industry. At the same time, interest in temporal databases has been increasing and a growing number of both prototype and implemented systems are using an enhanced temporal understanding to explain aspects of behavior associated with the implicit time-varying nature of the universe. This paper investigates the confluence of these two areas, surveys the work to date, and explores the issues involved and the outstanding problems in temporal data mining.

Modern organizations are accumulating huge volumes of textual documents. To turn archives into valuable knowledge sources, textual content must become explicit and able to be queried. Semantic tagging with markup languages such as XML satisfies both requirements. We thus introduce the DIAsDEM* framework for extracting semantics from structural text units (e.g., sentences), assigning XML tags to them and deriving a flat XML DTD for the archive. DIAsDEM focuses on archives characterized by a peculiar terminology and by an implicit structure such as court filings and company reports. In the knowledge discovery phase, text units are iteratively clustered by similarity of their content. Each iteration outputs clusters satisfying a set of quality criteria. Text units contained in these clusters are tagged with semiautomatically determined cluster labels and XML tags respectively. Additionally, extracted named entities (e.g., persons) serve as attributes of XML tags. We apply the framework in a case study on the German Commercial Register.

No abstract available.

Providers of Web based services are interested in monitoring the usage of their services in combination with those of other providers. The identification of services frequently accessed together may be valuable as a basis for strategic collaboration among their owners. We propose data mining to discover services of different providers which could complement one another, based on their usage. In particular we model the activities of a user as a sequence of service invocations recorded in a log, on which pattern discovery techniques can be applied. However we claim that conventional sequence mining is not adequate for this type of application. This is because, conventional mining concentrates on frequent (or infrequent) patterns of access, while we also require a notion of the consistency of these access patterns as a basis for collaboration. We present a model for constructing patterns that depict consistently used sequences of activities. This model is general enough to be applied to any system of autonomous entities, where relationships between entities are dynamic. For testing our model, we have analyzed the behavior of users in a news group, in order to determine consistent patterns in the way users respond to questions posed to the group.

Proposes a parallel optimizer for queries containing a large number of joins, as well as set operators and aggregate functions. The platform for the execution is a shared-disk multiprocessor machine supporting bushy parallelism and pipeline processing. Our model partitions the query into almost independent subtrees that can be optimized simultaneously, and it applies an enhanced variation of the iterative improvement technique on those subtrees which contain a large number of joins; this technique is parallelized, too. In order to estimate the cost of the states constructed during the optimization of join subtrees, cost formulae are developed that estimate the cost of relational algebra operators when executed across coalescing pipes.