Domain Expert Knowledge Research Articles

Paying attention to astronomical transients: introducing the time-series transformer for photometric classification

Abstract Future surveys such as the Legacy Survey of Space and Time (LSST) of the Vera C. Rubin Observatory will observe an order of magnitude more astrophysical transient events than any previous survey before. With this deluge of photometric data, it will be impossible for all such events to be classified by humans alone. Recent efforts have sought to leverage machine learning methods to tackle the challenge of astronomical transient classification, with ever improving success. Transformers are a recently developed deep learning architecture, first proposed for natural language processing, that have shown a great deal of recent success. In this work, we develop a new transformer architecture, which uses multihead self-attention at its core, for general multivariate time-series data. Furthermore, the proposed time-series transformer architecture supports the inclusion of an arbitrary number of additional features, while also offering interpretability. We apply the time-series transformer to the task of photometric classification, minimizing the reliance of expert domain knowledge for feature selection, while achieving results comparable with state-of-the-art photometric classification methods. We achieve a logarithmic-loss of 0.507 on imbalanced data in a representative setting using data from the Photometric LSST Astronomical Time-Series Classification Challenge. Moreover, we achieve a micro-averaged receiver-operating-characteristic area under curve of 0.98 and micro-averaged precision–recall area under curve of 0.87.

Trace encoding in process mining: A survey and benchmarking

Encoding methods are employed across several process mining tasks, including predictive process monitoring, anomalous case detection, trace clustering, etc. These methods are usually performed as preprocessing steps and are responsible for mapping complex event data information into a numerical feature space. Most papers choose existing encoding methods arbitrarily or employ a strategy based on expert domain knowledge. Moreover, existing methods are employed by using their default parameters without evaluating other options. This practice can lead to several drawbacks, such as suboptimal performance and unfair comparisons with the state-of-the-art. Therefore, this work aims at providing a comprehensive survey and benchmark on event log encoding by comparing 27 methods, from different natures, in terms of expressivity, scalability, correlation, and domain agnosticism. To the best of our knowledge, this is the most comprehensive study so far focusing on trace encoding in process mining. It contributes to maturing awareness about the role of trace encoding in process mining pipelines and sheds light on issues, concerns, and future research directions regarding the use of encoding methods to bridge the gap between machine learning models and process mining.

Data Standards and Interoperability Challenges for Biodiversity Digital Twin: A novel and transformative approach to biodiversity research and application

The Biodiversity Digital Twin (BioDT) project (2022-2025) aims to create prototypes that integrate various data sets, models, and expert domain knowledge enabling prediction capabilities and decision-making support for critical issues in biodiversity dynamics. While digital twin concepts have been applied in industries for continuous monitoring of physical phenomena, their application in biodiversity and environmental sciences presents novel challenges (Bauer et al. 2021, de Koning et al. 2023). In addition, successfully developing digital twins for biodiversity requires addressing interoperability challenges in data standards. BioDT is developing prototype digital twins based on use cases that span various data complexities, from point occurrence data to bioacoustics, covering nationwide forest states to specific communities and individual species. The project relies on FAIR principles (Findable, Accessible, Interoperable, and Reusable) and FAIR enabling resources like standards and vocabularies (Schultes et al. 2020) to enable the exchange, sharing, and reuse of biodiversity information, fostering collaboration among participating research infrastructures (DiSSCo, eLTER, GBIF, and LifeWatch) and data providers. It also involves creating a harmonised abstraction layer using Persistent Identifiers (PID) and FAIR Digital Object (FDO) records, alongside semantic mapping and crosswalk techniques to provide machine-actionable metadata (Schultes and Wittenburg 2019, Schwardmann 2020). Governance and engagement with research infrastructure stakeholders play crucial roles in this regard, with a focus on aligning technical and data standards discussions. In addition to data, models and workflows are key elements in BioDT. Models in the BioDT context are formal representations of problems or processes, implemented through equations, algorithms, or a combination of both, which can be executed by machine entities. The current twin prototypes are considering both statistical and mechanistic models, introducing significant variations in (1) data requirements, (2) modelling approaches and philosophy, and (3) model output. The BioDT consortium will develop guidelines and protocols for how to describe these models, what metadata to include, and how they will interact with the diverse datasets. While discussions on this topic exist within the broader context of biodiversity and ecological sciences (Jeltsch et al. 2013, Fer et al. 2020), the BioDT project is strongly committed to finding a solution within its scope. In the twinning context, data and models need to be executed within a computing infrastructure and also need to adhere to FAIR principles. Software within BioDT includes a suite of tools that facilitate data acquisition, storage, processing, and analysis. While some of these tools already exist, the challenge lies in integrating them within the digital twinning framework. One approach to achieving integration is through workflow representation, encompassing standardised procedures and protocols that guide the acquisition, packaging, processing, and analysis of data. The project is exploring Research Object Crate (RO-Crate) implementation for this (Soiland-Reyes et al. 2022). Implementing workflows can ensure reproducibility, scalability, and transparency in research practices, enabling scientists to validate and replicate findings. The BioDT project offers a novel and transformative approach to biodiversity research and application. By leveraging collaborative research infrastructures and adhering to data standards, BioDT aims to harness the power of data, software, supercomputers, models, and expertise to provide new insights. The foundation provided by the data standards, including those of Biodiversity Information Standards (TDWG), is crucial in realising the full potential of digital twins, facilitating the seamless integration of diverse data sources and combinations with models.

A label machine for mechanical systems: Discovering operating states with unsupervised learning from load time series

AbstractLabeling time series data according to operating states is often a time‐consuming task that requires expert domain knowledge of the underlying mechanical system. In this paper, we propose a data‐driven algorithm that identifies and detects operating states from time series data by grouping time ranges of similar signal behavior together using an unsupervised machine learning approach. The scattering transform and principal component analysis are utilized to extract signal characteristics from time series data which are subsequently clustered by a Gaussian mixture model to generate operating states. To evaluate our approach, we compare the automatically generated operating states with a manual definition of operating states created through expert knowledge. Based on a publicly available eBike dataset, the results demonstrate that the data‐driven definition of operating states can yield similar results to a rule set based on expert knowledge.

Real-Time Energy Management for Marine Applications Using Markov Approximation

All-electric and hybrid-electric ships have become the centerpiece for reducing greenhouse gas emissions and improving fuel efficiency in the maritime industry. Real-time power management of multiple power sources in both design and operation becomes critical due to the uncertainty and randomness of the vessel's power demand. Furthermore, the vessel operators may have additional requirements for operational flexibility, such as fully recharging the battery before the next trip and limiting engine switch on/off to reduce wear-and-tear, which lead to complex power management systems requiring domain-expert knowledge. In this paper, a hybrid solution for real-time power management is proposed. The model-based approach encapsulates the power system, and the data-driven approach (a machine learning technique called Markov approximation) handles the uncertainty. First, an offline deterministic optimization problem for power management is formulated. Second, the deterministic problem is transformed into a maximum weighted independent set (MWIS) problem. Next, the Markov approximation framework is applied to the transformed problem to utilize the machine learning techniques widely used in wireless communications and computing industries. Finally, a reinforcement learning-based real-time solution is proposed. Extensive simulations are performed on a ferry case study, and the results are within the proven theoretical bounds of machine learning techniques.

Evaluating the Use of Graph Neural Networks and Transfer Learning for Oral Bioavailability Prediction.

Oral bioavailability is a pharmacokinetic property that plays an important role in drug discovery. Recently developed computational models involve the use of molecular descriptors, fingerprints, and conventional machine-learning models. However, determining the type of molecular descriptors requires domain expert knowledge and time for feature selection. With the emergence of the graph neural network (GNN), models can be trained to automatically extract features that they deem important. In this article, we exploited the automatic feature selection of GNN to predict oral bioavailability. To enhance the prediction performance of GNN, we utilized transfer learning by pre-training a model to predict solubility and obtained a final average accuracy of 0.797, an F1 score of 0.840, and an AUC-ROC of 0.867, which outperformed previous studies on predicting oral bioavailability with the same test data set.

Breast cancer classification using deep learned features boosted with handcrafted features

Breast cancer is one of the leading causes of death among women across the globe. It is difficult to treat if detected at advanced stages. However, early detection can significantly increase chances of survival and improves the lives of millions of women. Given the widespread prevalence of breast cancer, it is of utmost importance for the research community to provide comprehensive framework encompassing early detection, classification, and diagnosis. The artificial intelligence research community, in coordination with medical practitioners, is developing such frameworks to automate the task of detection. With the surge in research activities coupled with the availability of large datasets and enhanced computational powers, it is expected that AI framework results will help even more clinicians in making correct predictions. In this article, a novel framework for the classification of breast cancer using mammograms is proposed. The proposed framework not only uses features extracted from Convolutional Neural Network (CNN) but also combines these features with handcrafted features (Histogram of Oriented Gradients (HOG) and Local Binary Pattern (LBP)) which helps in embedding domain expert knowledge, a step towards making proposed framework transparent. Experimental results conducted on the CBIS-DDSM dataset demonstrate that the proposed framework outperforms the current state-of-the-art methods in breast cancer classification.

A data fusion approach with mobile phone data for updating travel survey-based mode split estimates

Up-to-date information on different modes of travel to monitor transport traffic and evaluate rapid urban transport planning interventions is often lacking. Transport systems typically rely on traditional data sources providing outdated mode-of-travel data due to their data latency, infrequent data collection and high cost. To address this issue, we propose a method that leverages mobile phone data as a cost-effective and rich source of geospatial information to capture current human mobility patterns at unprecedented spatiotemporal resolution. Our approach employs mobile phone application usage traces to infer modes of transportation that are challenging to identify (bikes and ride-hailing/taxi services) based on mobile phone location data. Using data fusion and matrix factorisation techniques, we integrate official data sources (household surveys and census data) with mobile phone application usage data. This integration enables us to reconstruct the official data and create an updated dataset that incorporates insights from digital footprint data from application usage. We illustrate our method using a case study focused on Santiago, Chile successfully inferring four modes of transportation: mass-transit (all public transportation), motorised (cars and similar vehicles), active (pedestrian and cycle trips), and taxi (traditional taxi and ride-hailing services). Our analysis revealed significant changes in transportation patterns between 2012 and 2020. We quantify a reduction in mass-transit usage across municipalities in Santiago, except where metro/rail lines have been more recently introduced, highlighting added resilience to the public transport network of these infrastructure enhancements. Additionally, we evidence an overall increase in motorised transport throughout Santiago, revealing persistent challenges in promoting urban sustainable transportation. Findings also point to a rise in the share of taxi usage, and a drop in active mobility, suggesting a modal shift towards less sustainable modes of travel. We validate our findings comparing our updated estimates with official smart card transaction data. The consistency of findings with expert domain knowledge from the literature and historical transport usage trends further support the robustness of our approach.

Core–shell clustering approach for detection and analysis of coastal upwelling

A comprehensive approach is presented to analyze season’s coastal upwelling represented by weekly sea surface temperature (SST) image grids. The proposed model, core–shell clustering, assumes that the season’s upwelling can be divided into shorter periods of stability, time ranges, consisting of constant core and variable shell parts. A one-by-one core–shell clustering algorithm is provided. The algorithm parameters are automatically derived from the least-squares clustering criterion.The approach applies to SST gridded data for sixteen successive years (2004–2019) of coastal upwelling in the western Iberian coast, the northernmost branch of the Canary Current Upwelling System. Our results show that at each season, there are 3 to 5 time intervals, the ranges, at which the upwelling presents stable core patterns of relatively cold water surrounded by somewhat larger shell areas of warmer waters. Based on other experimental computations performed by our team, we conclude that this pattern is not just a purely local phenomenon but has a more global meaning. Inter-annual time series analysis are consistent among themselves and with existing expert domain knowledge.

A framework for enhancing industrial soft sensor learning models

Refinery industrial processes are very complex with nonlinear dynamics resulting from varying feedstock characteristics and also from changes in product prioritization. Along these processes, there are key properties of intermediate compounds that must be monitored and controlled since they directly affect the quality of the end products commercialized by these manufacturers. However, most of these properties can only be measured through time-consuming and expensive laboratory analysis, which is impossible to obtain in high frequencies, as required to properly monitor them. In this sense, developing soft sensors is the most common way to obtain high-frequency estimations for these measurements, helping advanced control systems to establish the correct setpoints for temperatures, pressures, and other sensors along the refining process, controlling the quality of end products. Since the amount of labeled data is scarce, most academic research has focused on employing semi- supervised learning strategies to develop machine learning (ML) models as soft sensors. Our research, on the other hand, goes in another direction. We aim to elaborate a framework that leverages the knowledge of domain experts and employs data augmentation techniques to build an enhanced fully labeled dataset that could be fed to any supervised ML algorithm to generate a quality soft sensor. We applied our framework together with Automated ML to train a model capable of predicting a specific key property associated with the production of Naphtha compounds in a refinery: the ASTM 95% distillation temperature of the Heavy Naphtha. Although our framework is model agnostic, we opted by using Automated ML for the optimization strategy, since it applies a diverse set of models to the dataset, reducing the bias of utilizing a single optimization algorithm. We evaluated the proposed framework on a case study carried out in an industrial refinery in Brazil, where the previous model in production for estimating the ASTM 95% distillation temperature of the Heavy Naphtha was based entirely on the physicochemical knowledge of the process. By adopting our framework with Automated ML, we were capable of improving the R2 score by 120%. The resulting ML model is currently operating in real-time inside the refinery, leading to significant economic gains.

Fault detection and diagnostics in the context of sparse multimodal data and expert knowledge assistance: Application to hydrogenerators

Deep learning-based methods for industrial fault detection and diagnostics (FDD) depend strictly on good quality and sufficient quantity of condition monitoring data. However, in real-world industrial settings, data collection is usually limited, leading to sparse and insufficient data to train a data-driven model. Therefore, this work proposes a new methodology to address this issue by leveraging multimodal data anddomain knowledge to develop a data-driven solution. Particularly for large, complex machinery, unimodal sensors may not fully capture the health state information. In such cases, multimodal data may provide complementary insights into the machine degradation. However, challenges mentioned above need to be addressed before these data can be useful. The multimodal learning method presented within the methodology can benefit from useful information from different data modalities and from domain expert knowledge, even when these data are of low volume. The performance of the proposed methodology is investigated through a real industrial case study involving energy production systems. The obtained results demonstrate the potential of the proposed methodology in augmenting the FDD accuracy and tackling the sparse data challenge.

Towards improving prediction accuracy and user-level explainability using deep learning and knowledge graphs: A study on cassava disease

Food security is currently a major concern due to the growing global population, the exponential increase in food demand, the deterioration of soil quality, the occurrence of numerous diseases, and the effects of climate change on crop yield. Sustainable agriculture is necessary to solve this food security challenge. Disruptive technologies, such as of artificial intelligence, especially, deep learning techniques can contribute to agricultural sustainability. For example, applying deep learning techniques for early disease classification allows us to take timely action, thereby helping to increase the yield without inflicting unnecessary environmental damage, such as excessive use of fertilisers or pesticides. Several studies have been conducted on agricultural sustainability using deep learning techniques and also semantic web technologies such as ontologies and knowledge graphs. However, the three major challenges remain: (i) the lack of explainability of deep learning-based systems (e.g. disease information), especially to non-experts like farmers; (ii) a lack of contextual information (e.g. soil or plant information) and domain-expert knowledge in deep learning-based systems; and (iii) the lack of pattern learning ability of systems based on the semantic web, despite their ability to incorporate domain knowledge. Therefore, this paper presents the work on disease classification, addressing the challenges as mentioned earlier by combining deep learning and semantic web technologies, namely ontologies and knowledge graphs. The findings are: (i) 0.905 (90.5%) prediction accuracy on large noisy dataset; (ii) ability to generate user-level explanations about disease and incorporate contextual and domain knowledge; (iii) the average prediction latency of 3.8514 s on 5268 samples; (iv) 95% of users finding the explanation of the proposed method useful; and (v) 85% of users being able to understand generated explanations easily—show that the proposed method is superior to the state-of-the-art in terms of performance and explainability and is also suitable for real-world scenarios.

Knowledge-Graph- and GCN-Based Domain Chinese Long Text Classification Method

In order to solve the current problems in domain long text classification tasks, namely, the long length of a document, which makes it difficult for the model to capture key information, and the lack of expert domain knowledge, which leads to insufficient classification accuracy, a domain long text classification model based on a knowledge graph and a graph convolutional neural network is proposed. BERT is used to encode the text, and each word’s corresponding vector is used as a node for the graph convolutional neural network so that the initialized vector contains rich semantic information. Using the trained entity–relationship extraction model, the entity-to-entity–relationships in the document are extracted and used as the edges of the graph convolutional neural network, together with syntactic dependency information. The graph structure mask is used to learn about edge relationships and edge types to further enhance the learning ability of the model for semantic dependencies between words. The method further improves the accuracy of domain long text classification by fusing knowledge features and data features. Experiments on three long text classification datasets—IFLYTEK, THUCNews, and the Chinese corpus of Fudan University—show accuracy improvements of 8.8%, 3.6%, and 2.6%, respectively, relative to the BERT model.

A Study on Factors Associated with Child Sexual Abuse and Recognizing the Severity: Special Reference to Galle District

Child Sexual Abuse has been a global epidemic with devastating consequences. One in four girls and one in six boys have been experienced some form of sexual abuse in their tender age in the world. According to Police statistics, Child Sexual Abuse (CSA) cases is growing in recent years in Sri Lanka too. Galle is among the four districts where the reported child abuse cases high and the reported CSA complaints are increasing extraordinarily. Also, there is no previous research have been done in the Southern part of the country regarding the crisis of CSA. So, main objective of this study is to determine the key risk factors that affected to a CSA in Galle Police Division, and to develop suitable regression and machine learning models to predict the severity of CSA. 225 CSA cases reported to Police Child and Women Bureau of Galle Police Division during the period 2017 – 2020 were treated for this study. Out of twenty-one risk which were found from literature and knowledge of domain experts, sixteen variables showed a significant relationship with response variable severity of CSA according to chi-square test of association. Traditional OLR model was performed to predict severity of CSA and to detect key risk factors to a CSA with two different data selection methods. Next, machine learning techniques: Decision Tree, SVM, and PNN were trained to classify severity of CSA. Random over-sampling technique was used to overcome the class imbalanced problem persists in the dataset. Finally, bagging technique was executed to conserve robustness of models and to improve performance. The OLR model classified the severity of CSA with 68.85% accuracy. Machine learning techniques, Decision Tree, SVM and PNN model classified the severity of CSA with an accuracy of 82.15%, 77.68% and 85.25% respectively. PNN model performed with higher accuracy better than other fitted models. The results obtained from this study can be used to take precautions and to arrange awareness sessions for adults to reduce CSA in Galle Police Division. Also, the study can be extended to the whole island to reduce CSA and to make it a better place for children.

Beyond Low-Pass Filtering: Graph Convolutional Networks With Automatic Filtering

Graph convolutional networks are becoming indispensable for deep learning from graph-structured data. Most of the existing graph convolutional networks share two big shortcomings. First, they are essentially low-pass filters, thus the potentially useful middle and high frequency band of graph signals are ignored. Second, the bandwidth of existing graph convolutional filters is fixed. Parameters of a graph convolutional filter only transform the graph inputs without changing the curvature of a graph convolutional filter function. In reality, we are uncertain about whether we should retain or cut off the frequency at a certain point unless we have expert domain knowledge. In this paper, we propose Automatic Graph Convolutional Networks (AutoGCN) to capture the full spectrum of graph signals and automatically update the bandwidth of graph convolutional filters. While it is based on graph spectral theory, our AutoGCN is also localized in space and has a spatial form. Experimental results show that AutoGCN achieves significant improvement over baseline methods which only work as low-pass filters.

Comparison of two artificial intelligence-augmented ECG approaches: Machine learning and deep learning.

Artificial intelligence-augmented ECG (AI-ECG) refers to the application of novel AI solutions for complex ECG interpretation tasks. A broad variety of AI-ECG approaches exist, each having differing advantages and limitations relating to their creation and application. To provide illustrative comparison of two general AI-ECG modeling approaches: machine learning (ML) and deep learning (DL). Two AI-ECG algorithms were developed to carry out two separate tasks using ML and DL, respectively. ML modeling techniques were used to create algorithms designed for automatic wide QRS complex tachycardia differentiation into ventricular tachycardia and supraventricular tachycardia. A DL algorithm was formulated for the task of comprehensive 12‑lead ECG interpretation. First, we describe the ML models for WCT differentiation, which rely upon expert domain knowledge to identify and formulate ECG features (e.g., percent monophasic time-voltage area [PMonoTVA]) that enable strong diagnostic performance. Second, we describe the DL method for comprehensive 12‑lead ECG interpretation, which relies upon the independent recognition and analysis of a virtually incalculable number of ECG features from a vast collection of standard 12‑lead ECGs. We have showcased two different AI-ECG methods, namely ML and DL respectively. In doing so, we highlighted the strengths and weaknesses of each approach. It is essential for investigators to understand these differences when attempting to create and apply novel AI-ECG solutions.

RoSGAS : Adaptive Social Bot Detection with Reinforced Self-supervised GNN Architecture Search

Social bots are referred to as the automated accounts on social networks that make attempts to behave like humans. While Graph Neural Networks (GNNs) have been massively applied to the field of social bot detection, a huge amount of domain expertise and prior knowledge is heavily engaged in the state-of-the-art approaches to design a dedicated neural network architecture for a specific classification task. Involving oversized nodes and network layers in the model design, however, usually causes the over-smoothing problem and the lack of embedding discrimination. In this article, we propose RoSGAS , a novel R einf o rced and S elf-supervised G NN A rchitecture S earch framework to adaptively pinpoint the most suitable multi-hop neighborhood and the number of layers in the GNN architecture. More specifically, we consider the social bot detection problem as a user-centric subgraph embedding and classification task. We exploit the heterogeneous information network to present the user connectivity by leveraging account metadata, relationships, behavioral features, and content features. RoSGAS uses a multi-agent deep reinforcement learning (RL), 31 pages. mechanism for navigating the search of optimal neighborhood and network layers to learn individually the subgraph embedding for each target user. A nearest neighbor mechanism is developed for accelerating the RL training process, and RoSGAS can learn more discriminative subgraph embedding with the aid of self-supervised learning. Experiments on five Twitter datasets show that RoSGAS outperforms the state-of-the-art approaches in terms of accuracy, training efficiency, and stability and has better generalization when handling unseen samples.

Collaboration between instructional designers and subject matter experts in digital transformation projects

The rapid adoption of online learning during the Covid-19 pandemic and digital transformation associated with new tools and technologies has driven the demand for interprofessional collaboration within the education sector. It is now commonplace for Instructional Designers (IDs) and Learning Technologists (LTs) to work alongside Subject Matter Experts (SMEs) when creating online learning content. Although the professional practice of an ID may vary, the collaborative approach to working with SMEs is similar across sectors. Typically, the role of the SME is to impart expert domain knowledge, skills or behaviours. An ID captures this information and creates a learning experience using pedagogical expertise and design skills. This relationship is not without friction and can be complex and challenging to navigate. Research often focuses on the experiences of SMEs and suggests a simple resolution to avoid conflict; clearly define each agent’s role in writing. However, this is not a clear-cut solution. Detailed guidance for IDs is lacking, and conflicts still exist. This study aimed to gain a deeper insight into the interprofessional working relationship between IDs and SMEs. A series of semi-structured interviews were conducted, which looked through the lens of relational working theory and focused on the perspective of the ID. Key themes were identified and aligned with the core concepts of the relational working framework. Findings show that a communication breakdown can negatively impact a project, resulting in power struggles and the delivery of subpar content. The discussion highlights how the application of the relational working framework could be used to strengthen and ease tensions between SMEs and IDs.Keywords: instructional designers; subject matter experts; relational working theory; relational working framework; digital transformationPart of the Special Issue Teaching practices in times of digital transformation <https://doi.org/10.21428/8c225f6e.ac86609b>

Systematic reviews as a metaknowledge tool: caveats and a review of available options

AbstractThe development of information and communication technologies accompanied by the exponential growth of electronic scholarly publications and computer archives has led, on the one hand, to the development of techniques for the systematic analysis of published literature and, on the other hand, to a significant growth in the number of literature reviews conducted and published. A systematic literature review seeks to comprehensively identify and synthesize all studies conducted on a given topic. However, how does one ensure that the entire state of the art on a given topic has been covered? How does one collect and select a large number of publications on a given topic, and how does one analyze the knowledge extracted from these publications? In other words, how do we use systematic literature analysis to produce new knowledge from existing knowledge? To make the best use of the various tools available today for conducting systematic reviews, it is necessary to be aware of their limitations. In the paper, we provide illustrations, examples, and detailed information on the various alternatives for conducting systematic reviews, presenting a review of available sources and tools. We report an analysis of the articles published in the journal International Transactions in Operational Research to show the potential and limitations of systematic review versus domain expert‐based selection. Finally, we show that a good systematic review relies on the intelligent combination of the latest tools available for conducting systematic research with the knowledge of domain experts.

Wearable-Based Intelligent Emotion Monitoring in Older Adults during Daily Life Activities

We present a system designed to monitor the well-being of older adults during their daily activities. To automatically detect and classify their emotional state, we collect physiological data through a wearable medical sensor. Ground truth data are obtained using a simple smartphone app that provides ecological momentary assessment (EMA), a method for repeatedly sampling people’s current experiences in real time in their natural environments. We are making the resulting dataset publicly available as a benchmark for future comparisons and methods. We are evaluating two feature selection methods to improve classification performance and proposing a feature set that augments and contrasts domain expert knowledge based on time-analysis features. The results demonstrate an improvement in classification accuracy when using the proposed feature selection methods. Furthermore, the feature set we present is better suited for predicting emotional states in a leave-one-day-out experimental setup, as it identifies more patterns.