The development of aberrant cells in skin tissues is generally related to skin cancer and cutaneous diseases. They occur as a result of damage to DNA cells, primarily from exposure to ultraviolet radiation from the Sun. These cells may be benign or malignant, in other words, non-cancerous or cancerous, respectively. Patients’ chances of recovery could be improved by the precise early diagnosis of malignant skin lesions. In that sense, dermatologists need to embrace the promise that artificial intelligence techniques present. In this paper, we apply and compare multiple state-of-the-art deep learning architectures to classify benign and malignant skin lesions using a comprehensive image dataset. In addition, we explore the potential of model fusion techniques, combining multiple architectures to enhance the classification accuracy. The results obtained are then compared to measure their effectiveness.

Quantum machine learning (QML) offers new perspectives for modeling complex sequential data by leveraging the computational advantages of quantum computing. In this work, we explore the application of Quantum Long Short-Term Memory (QLSTM) networks for sequence prediction tasks. By integrating parameterized quantum circuits within the classical LSTM architecture, QLSTM models exploit quantum parallelism and entanglement to learn temporal dependencies in data more efficiently. We evaluate the performance of QLSTM on synthetic and real-world sequence prediction tasks and compare it with its classical counterpart. The results demonstrate that QLSTM models achieve competitive performance with fewer trainable parameters, indicating their potential as lightweight and efficient models for sequence learning and prediction.

Competency tagging plays an essential role in both academic and industrial settings, enabling the alignment of learning content, job postings, and resumes with specific skill sets. However, traditional manual tagging is costly, time-intensive, and prone to inconsistencies. In this work, we propose an automated competency tagging method leveraging semantic search with fine-tuned Large Language Model (LLM). Our approach encodes textual data from learning materials and competency descriptions into a shared embedding space, enabling efficient retrieval of relevant competency tags via similarity search. We systematically evaluate semantic matching at different levels of granularity-document, paragraph, and sentence-to optimize retrieval performance. Furthermore, we fine-tune the LLM using Low-Rank Adaptation (LoRA) to improve competency tagging while maintaining efficiency. Experiments on a dataset of 164 pages of learning content and 96 competencies demonstrate the effectiveness of our method, achieving a recall@10 of $80.29 \%$. Notably, fine-tuning with LoRA led to a $6 \%$ improvement in recall@10, highlighting its impact on enhancing retrieval performance. Our findings underscore the potential of fine-tuned LLMs for high-precision competency tagging.

This paper proposes a new approach based on sentiment analysis for predicting the relevance of religious beliefs in contemporary relationships within the context of online dating. The motivation for this research lies in understanding how shared values, common missions, and rituals in various religions might influence decision-making in relationships, the study aims to predict users’ religious beliefs based on OkCupid profiles. Our approach starts first by 1) extracting relevant features using sentiment analysis from text data, and 2) developing machine learning models based on state-of-the-art algorithms such as random forests, logistic regression, and decision trees. The results obtained demonstrated the effectiveness of our proposed approach in accurately predicting religious beliefs and its superiority compared to the state-of-the-art methods. Through this investigation, the paper seeks insights into the interplay between religious beliefs and romantic choices.

Social media platforms assemble people to discuss various types of topics and to share information. Increasingly, the use of social media is exceeding the scope of its purposes. For instance, some people utilize social media to mobilize an important number of proponents to organize different types of demonstrations, political demands, and political movements and events. In this paper, we are interested in investigating the evolution of the Western exit (Wexit) movement in social media. Specifically, we jointly perform sentiment analysis and topic models to track over time the sentiment polarity of Wexit-related topics, and to discover ever-growing communities and the influence that they have over other communities. The experimental study on the Wexit movement data showed the performance of our proposed method and its suitability for tracking communities over time compared to state-of-the art methods.

The emergence of combining different Deep Learning architectures in the Artificial Intelligence research field to perform a specific task offers new perspectives and at times, performance enhancement. In this paper, multiple levels of fusion based on single Deep Learning and Ensemble Deep Learning models is introduced while tackling the national electrical energy forecasting task of Algeria on a short-term basis using the multi-sourced data related to load demand and meteorological factors, provided by the System Operator of Algerian National Electricity and Gas Company (SONELGAZ). The different model architectures are based on Stacked Denoising Auto-encoder and One Dimensional Convolutional Neural Network. The different approaches' efficiency is tested by comparing the different empirical results based on the Mean Absolute Percentage Error. The obtained results not only underline the efficiency and precision of deep ANNs architectures but also show the impact of the Ensemble Deep Learning method when incorporated with the right level of information fusion.

Anomaly detection is crucial for maintaining the reliability and security of wireless sensor networks and loT systems. Conventional methods require labeled data, often unavailable in these systems, making unsupervised techniques essential. Deep learning has shown promise in unsupervised anomaly detection tasks, including wireless sensor networks and loT applications. The Isolation Forest, an unsupervised anomaly detection algorithm, isolates anomalies based on in-herent properties. Combining deep learning with the Isolation Forest can improve accuracy and effectiveness in detecting anomalies. This paper presents the Deep Generative Model with Isolation Forest (DGM-IF), a novel unsupervised anomaly detection method for wireless sensor networks and loT. DGM-IF integrates deep generative models with the Isolation Forest algorithm to learn a robust representation of normal data and identify anomalies. The model generates synthetic data based on the learned distribution and employs the Isolation Forest to separate deviating data points. The proposed technique is assessed using real-world datasets and benchmarked against cutting-edge methods, proving its efficacy in detecting anomalies. The DGM-IF approach has the potential to significantly enhance the reliability and security of wireless sensor networks and IoT systems by identifying potential threats and attacks.

Abstract Machine learning (ML) algorithms have become popular in recent years and have found increasing utility in the field of medical imaging, specifically in positron emission tomography (PET) imaging. The interest in ML in PET imaging for the study of neurodegenerative diseases stems from the potential of these techniques to analyze and predict the physiological parameters of biomarkers such as the total volume of distribution (V $$_{\text {t}}$$ t ) in the organ or a structure of the organ to be explored. In this paper, we investigated whether the V $$_{\text {t}}$$ t of [ $$^{18}$$ 18 F]-FEPPA radiotracer, an indicator of neuroinflammation, could be estimated directly in a non-invasive way, given the activity of the radiotracer in brain tissue. The study used several regression models to predict the [ $$^{18}$$ 18 F]-FEPPA V $$_{\text {t}}$$ t in different brain regions where 31 regions of interest were defined for each of 24 patients with Parkinson disease and 20 healthy subjects, and were used to train four tree-based regression models. The predicted and reference values were compared by Bland-Altman analysis and regression model’s performance was evaluated by the mean absolute error (MAE). The best result was obtained by the XGBoost model with a MAE of 2.6. Bland-Altman analysis results indicate that predicted V $$_{\text {t}}$$ t are in average very close to the reference with a bias of 0.23 "Image missing" 2.82. Significant main effect of genotype on [ $$^{18}$$ 18 F]-FEPPA in both caudate and putamen have been preserved by predicted Vt values (p < 0.05). The results of paired t-test indicate that the difference between predicted and reference V $$_{\text {t}}$$ t is not statistically significant in 6 out of 8 groups. The proposed algorithms provide a non-invasive and efficient tool to predict [ $$^{18}$$ 18 F]-FEPPA V $$_{\text {t}}$$ t values, a hallmark of neuroinflammation that is believed to be a potential trigger for Parkinson’s disease development.

Abstract Children diagnosed with Autism Spectrum Disorder (ASD) often exhibit agitated behaviors that can isolate them from their peers. This study aims to examine if wearable data, collected during everyday activities, could effectively detect such behaviors. First, we used the Empatica E4 device to collect real data including Blood Volume Pulse (BVP), Electrodermal Activity (EDA), and Acceleration (ACC), from a 9-years-old male child with autism over 6 months. Second, we analyzed and extracted numerous features from each signal, and employed different classifiers including Support Vector Machine (SVM), Random Forest (FR), eXtreme Gradient Boosting (XGBoost), and TabNet. Our preliminary findings showed good performance in comparison with the state of the art. Notably, XGBoost demonstrated the highest performance in terms of accuracy, precision, recall, and F1-score. The accuracy achieved in this paper using XGBoost is equal to $$80\%$$ 80% which exceeds previous research.

Psychology research findings suggest that personality is related to differences in friendship characteristics and that some personality traits correlate with linguistic behavior. In this paper, we investigate the influence that personality may have on affinity formation. To this end, we derive affinity relationships from social media interactions, examine personality based on language use to discover the emotional stability of affinity relationships, and measure semantic similarity at the personality type level to understand the logic behind the development of affinity. Specifically, we conduct extensive experiments using a publicly available dataset containing information on individuals who self-identified with a Myers-Briggs personality type. Our results identify certain influential personality types that weigh more heavily on affinity relationships and show that personality can be predicted from spontaneous language with an F-1 score superior to 0.76. Future research avenues are proposed.

Understanding the factors that lead or contribute to emotional instability in highly motivated high-conflict dialogues such as law enforcement interviews can be of crucial importance. In this paper, we extract psycholinguistic features to assess emotional stability scale development and identify patterns that are relevant to emotional breakdown. To this end, we utilize zero-shot text classification to investigate the temporal evolution of emotion during law enforcement interviews. We conduct ex-tensive experiments using publicly available police interrogation transcripts. Our results are promising and suggest avenues for future research.

Abstract For medical treatments, pain is often measured by self-report. However, the current subjective pain assessment highly depends on the patient’s response and is therefore unreliable. In this paper, we propose a physiological-signals-based objective pain recognition method that can extract new features, which have never been discovered in pain detection, from electrodermal activity (EDA) and electrocardiogram (ECG) signals. To discriminate the absence and presence of pain, we establish four classification tasks and build four tree-based classifiers, including Random Forest, Adaptive Boosting (AdaBoost), eXtreme Gradient Boosting (XGBoost), and TabNet. The comparative experiments demonstrate that our method using the EDA and ECG features yields accurate classification results. Furthermore, the TabNet achieves a large accuracy improvement using our ECG features and a classification accuracy of 94.51% using the features selected from the fusion of the two signals.

Short-term load forecasting is a topic of considerable interest as it is of major importance for specifying and managing power resources and needs. In the literature, Deep Neural Networks have been successfully recently applied in load forecasting using single modalities as an improvement to traditional Artificial Neural Networks (ANN). In this paper, the main objective is to tackle the load forecasting problem with the intention of enhancing the prediction performance by combining and testing different multi-modal deep learning approaches and architectures in order to process and relate information from multiple modalities. The benefits of using multi-modality instead of one modality, applied to the time series modeling problem of short-term load forecasting is therefore investigated. The models are trained and evaluated using the hourly temperature and electrical consumption in addition to auto-regressive variables as the first modality. The day type characteristic, such as: weekends, week days, bank holidays, religious holidays etc., may be considered as a second modality. The approach liability is tested by comparing the empirical results of different Deep architectures namely: Stacked Denoising Auto- Encoders (SDAEs) and Convolutional Neural Network (CNN), with multiple and single modalities where processing the same task of one day ahead load forecasting.

Early detection of behavioral deterioration can be of great importance in preventing individuals' misbehavior from escalating in severity. This paper addresses the problem of behavioral deterioration in the context of online discussion forums. We propose a novel method that builds behavioral sequences from temporal information to gain a better understanding of behaviors exhibited by forum members, and then explores n-gram features to predict behavioral deterioration from consecutive combinations of sequential patterns corresponding to misbehavior. We conduct extensive experiments using real-world datasets and demonstrate the ability of our method to predict behavioral deterioration with a high degree of accuracy, as evaluated by F-1 scores. Our quantitative analysis of the model's performance yields F-1 scores of over 0.7. Specifically, we find that the best-performing model is linear SVM, with an average F-1 score of 0.74. Some future research avenues are proposed.

Stance detection is the task of automatically determining whether the author of a piece of text is in favor of, against, or neutral towards a target such as a topic, entity, or claim. In this paper, we propose a method based on RoBERTa to classify stances by capturing the context of the discussion through the examination of pairs of stances and relational structures of debates specific to each topic within the defined window of each forum participant's interventions. Furthermore, we examine the degree of disagreement and neutrality in various debate topics to measure divergence of opinion in the course of the debate and estimate the emotional state manifested in different debate topics. We conduct extensive experiments using two publicly available datasets and demonstrate that our method considers more stance classes, provides better results and yields statistical improvements over existing techniques. Our quantitative analysis of model performance yields F-1 scores of over 0.745. Interestingly, we obtained the highest F-1 score, 0.814, on a stance class which was not taken into consideration in prior work. We report that none of the metrics utilized to measure divergence of opinion yield values exceeding 50 % and the correlations between the same topics over 10-fold cross-validation are statistically significant for the majority of them (p <; 0.005). Several future research avenues are proposed.

Short-term load forecasting is a topic of considerable interest; it ensures the balance between the production and consumption one day ahead. In this paper, time series models have been developed to provide an efficient forecast for electricity consumption in Algeria using Deep Neural Networks in the form of Stacked Denoising Autoencoder (SDAE) and a regular Multilayer Perceptron (MLP) as a benchmark model. The obtained models are established and evaluated using the hourly temperature and electricity consumption data provided by the Algerian National Electricity and Gas Company (SONELGAZ). Convincing forecasting results for the Algerian national load were found and conclusions drawn.

This paper addresses the problem of discovering hidden affinity relationships in online communities. Online discussions assemble people to talk about various types of topics and to share information. People progressively develop the affinity, and they get closer as frequently as they mention themselves in messages and they send positive messages to one another. We propose an algorithm, named HAR-search, for discovering hidden affinity relationships between individuals. Based on Markov Chain Models, we derive the affinity scores amongst individuals in an online community. We show that our method allows to track the evolution of the affinity over time and to predict affinity relationships arisen from the influence of certain community members. The comparison with the state-of-the-art method shows that our method results in robust discovery and considers minute details.

This paper addresses a new problem of automatic detection of visual attention in older adults based on their driving speed. All state-of-the-art methods try to understand the on-road performance of older adults by means of the Useful Field of View (UFOV) measure. Our method takes advantage of deep learning models such as Long-short Term Memory (LSTM) to automatically extract features from driving speed data for predicting drivers’ visual attention. We demonstrate, through extensive experiments on real dataset, that our method is able to predict the driver’s visual attention based on driving speed with high accuracy.

Smart home refers to the independency and comfort that are ensured by remote monitoring and assistive services. Assisting an elderly person requires identifying and accurately predicting his/her normal and abnormal behaviors. Abnormal behaviors observed during the completion of activities of daily living are a good indicator that the person is more likely to have health and behavioral problems that need intervention and assistance. In this paper, we propose a method, based on long short-term memory recurrent neural networks (LSTM), to automatically predicting an elderly person’s abnormal behaviors. Our method allows to model the temporal information expressed in the long sequences collected over time. Our study aims to evaluate the performance of LSTM on identifying and predicting elderly persons abnormal behaviors in smart homes. We experimentally demonstrated, through extensive experiments using a dataset, the suitability and performance of the proposed method in predicting abnormal behaviors with high accuracy. We also demonstrated the superiority of the proposed method compared to the existing state-of-the-art methods.

Users in real-world social networks are organized into communities that differ from each other in terms of influence, authority, interest, size, etc. This article addresses the problems of detecting communities of authority and of estimating the influence of such communities in dynamic social networks. These are new issues that have not yet been addressed in the literature, and they are important in applications such as marketing and recommender systems. To facilitate the identification of communities of authority, our approach first detects communities sharing common interests, which we call “meta-communities,” by incorporating topic modeling based on users’ community memberships. Then, communities of authority are extracted with respect to each meta-community, using a new measure based on the betweenness centrality. To assess the influence between communities over time, we propose a new model based on the Granger causality method. Through extensive experiments on a variety of social network datasets, we empirically demonstrate the suitability of our approach for community-of-authority detection and assessment of the influence between communities over time.

This paper addresses a new problem concerning the evolution of influence relationships between communities in dynamic social networks. A weighted temporal multigraph is employed to represent the dynamics of the social networks and analyze the influence relationships between communities over time. To ensure the interpretability of the knowledge discovered, evolution of the influence relationships is assessed by introducing the Granger causality. Through extensive experiments, we empirically demonstrate the suitability of our model for studying the evolution of influence between communities. Moreover, we empirically show how our model is able to accurately predict the influence of communities over time using random forest regression.

This paper addresses a new problem concerning the discovery and tracking of influencer-influencee relationships between communities in dynamic social networks. A weighted temporal multigraph is employed to represent the dynamics of the social networks. To discover and track influencer-influencee relationships over time, communities sharing common interests are first grouped together in meta-communities using a topic modeling approach. Then, influencer-influencee relationships are discovered and tracked using the transfer entropy causality method. Through extensive experiments on the DBLP research publication dataset, we empirically demonstrate the suitability of our model for the discovery of influencer-influencee relationships between communities and the tracking of such relationships over time.

This paper presents a novel model for user profile construction using causal relationships. Causal relationships are extracted from behavior sequences to build user profiles. Our model first discovers significant patterns by adapting a new sequence clustering algorithm, and then discovers pattern associations using normalized mutual information (NMI). Causal relationships between significant patterns are then extracted using the transfer entropy approach. These relationships are used to construct causal graphs of activities, to generate the user profile. In extensive experiments on a variety of datasets, we empirically demonstrate that these causality-based profiles yield a significant increase in performance on activity prediction.

This paper presents a practical and novel model for behavioral user profile construction using causal relationships. Causal relationships are extracted from behavior sequences for building user profiles. Our model discovers significant patterns from behavior sequences, then it discovers patterns associations using normalized mutual information. Causal relationships between significant patterns are then identified using the transfer entropy approach. We empirically demonstrate that these causality-based profiles accurately describe users profiles and allow developing practical Ubicomp applications.

This paper presents an approach for recognition of Activities of Daily Living (ADLs) in smart environments. Our approach is based on the frequent pattern mining principle to extract frequent patterns in the datasets collected from different sensors disseminated in a smart environment. In contrast with existing intrusive activity recognition approaches that have been proposed in the literature, where the datasets are basically composed of audio-visual or images files recorded during experiments, our approach is fully non-intrusive and it is based on the analysis of event sequences collected from heterogenous sensors. Our approach consists of two main phases, (1) frequent pattern mining to extract frequent patterns, and (2) activity recognition using a mapping function between the extracted frequent patterns and the activity models. We show through experiments how our approach accurately recognizes tasks as well as activities and outperforms the HMM model.

Designing human machine interfaces that respect the ergonomic norms and following rigorous approaches constitutes a major concern for computer systems designers. The increased need on easily accessible and usable interfaces leads researchers in this domain to create methods and models that make it possible to evaluate these interfaces in terms of utility and usability. Two different approaches are currently used to evaluate human machine interfaces, empirical approaches that require user involvement in the interface development process, and analytical approaches that do not associate the user during the interface development process. This paper presents a study of user performance on two principal tasks of the contextual assistant’s interface, developed in the context of smart homes, to assist persons with cognitive disabilities. We use three different methods to analyze and evaluate this interface, focusing basically on time of execution. Two of the models developed are based on cognitive models, which are ACT-R and GOMS and the third one is based on the Fitts’ Law model. The results show that, all models give a good prediction of user performance, even if the cognitive models show better accuracy of the user performance. Furthermore, they provide a better insight into cognitive abilities required to interact with the interface.

The growth of ubiquitous computing and its integration in smart environments constitutes a major challenge. However, many issues related to this technology were raised, and many efforts have been made to facilitate the implementation of such technology. Information management, sharing and reuse are examples of these issues. Ontologies are used in this context as promising solution to deal with these issues. Nevertheless, ontologies developed to date in the context of smart environments are limited and do not address all aspects of smart environments such as the problem of referentiality and the environmental change. In this paper, we propose a universal ontology for smart environments aiming firstly to overcome the limitations of the existing ontologies, and secondly extend its capabilities by adding new environmental aspects such as those mentioned previously. In this paper, we also demonstrate how our ontology can be used to describe domains as well as applications.