Your search

Artificial intelligence (AI) and deep learning (DL) are advancing in stock market prediction, attracting the attention of researchers in computer science and finance. This bibliometric review analyzes 525 articles published from 1991 to 2024 in Scopus-indexed journals, utilizing VOSviewer software to identify key research trends, influential contributors, and burgeoning themes. The bibliometric analysis encompasses a performance analysis of the most prominent scientific contributors and a network analysis of scientific mapping, which includes co-authorship, co-occurrence, citation, bibliographical coupling, and co-citation analyses enabled by the VOSviewer software. Among the 693 countries, significant hubs of knowledge production include China, the US, India, and the UK, highlighting the global relevance of the field. Various AI and DL technologies are increasingly employed in stock price predictions, with artificial neural networks (ANN) and other methods such as long short-term memory (LSTM), Random Forest, Sentiment Analysis, Support Vector Machine/Regression (SVM/SVR), among the 1399 keyword counts in publications. Influential studies such as LeBaron (1999) and Moghaddam (2016) have shaped foundational research in 8159 citations. This review offers original insights into the bibliometric landscape of AI and DL applications in finance by mapping global knowledge production and identifying critical AI methods advancing stock market prediction. It enables finance professionals to learn about technological developments and trends to enhance decision-making and gain market advantage.

Even with more than 12 billion vaccine doses administered globally, the Covid-19 pandemic has caused several global economic, social, environmental, and healthcare impacts. Computer Aided Diagnostic (CAD) systems can serve as a complementary method to aid doctors in identifying regions of interest in images and help detect diseases. In addition, these systems can help doctors analyze the status of the disease and check for their progress or regression. To analyze the viability of using CNNs for differentiating Covid-19 CT positive images from Covid-19 CT negative images, we used a dataset collected by Union Hospital (HUST-UH) and Liyuan Hospital (HUST-LH) and made available at the Kaggle platform. The main objective of this chapter is to present results from applying two state-of-the-art CNNs on a Covid-19 CT Scan images database to evaluate the possibility of differentiating images with imaging features associated with Covid-19 pneumonia from images with imaging features irrelevant to Covid-19 pneumonia. Two pre-trained neural networks, ResNet50 and MobileNet, were fine-tuned for the datasets under analysis. Both CNNs obtained promising results, with the ResNet50 network achieving a Precision of 0.97, a Recall of 0.96, an F1-score of 0.96, and 39 false negatives. The MobileNet classifier obtained a Precision of 0.94, a Recall of 0.94, an F1-score of 0.94, and a total of 20 false negatives.

The gold standard to detect SARS-CoV-2 infection considers testing methods based on Polymerase Chain Reaction (PCR). Still, the time necessary to confirm patient infection can be lengthy, and the process is expensive. In parallel, X-Ray and CT scans play an important role in the diagnosis and treatment processes. Hence, a trusted automated technique for identifying and quantifying the infected lung regions would be advantageous. Chest X-rays are two-dimensional images of the patient’s chest and provide lung morphological information and other characteristics, like ground-glass opacities (GGO), horizontal linear opacities, or consolidations, which are typical characteristics of pneumonia caused by COVID-19. This chapter presents an AI-based system using multiple Transfer Learning models for COVID-19 classification using Chest X-Rays. In our experimental design, all the classifiers demonstrated satisfactory accuracy, precision, recall, and specificity performance. On the one hand, the Mobilenet architecture outperformed the other CNNs, achieving excellent results for the evaluated metrics. On the other hand, Squeezenet presented a regular result in terms of recall. In medical diagnosis, false negatives can be particularly harmful because a false negative can lead to patients being incorrectly diagnosed as healthy. These results suggest that our Deep Learning classifiers can accurately classify X-ray exams as normal or indicative of COVID-19 with high confidence.

The Covid-19 pandemic evidenced the need Computer Aided Diagnostic Systems to analyze medical images, such as CT and MRI scans and X-rays, to assist specialists in disease diagnosis. CAD systems have been shown to be effective at detecting COVID-19 in chest X-ray and CT images, with some studies reporting high levels of accuracy and sensitivity. Moreover, it can also detect some diseases in patients who may not have symptoms, preventing the spread of the virus. There are some types of CAD systems, such as Machine and Deep Learning-based and Transfer learning-based. This chapter proposes a pipeline for feature extraction and classification of Covid-19 in X-ray images using transfer learning for feature extraction with VGG-16 CNN and machine learning classifiers. Five classifiers were evaluated: Accuracy, Specificity, Sensitivity, Geometric mean, and Area under the curve. The SVM Classifier presented the best performance metrics for Covid-19 classification, achieving 90% accuracy, 97.5% of Specificity, 82.5% of Sensitivity, 89.6% of Geometric mean, and 90% for the AUC metric. On the other hand, the Nearest Centroid (NC) classifier presented poor sensitivity and geometric mean results, achieving 33.9% and 54.07%, respectively.

Association Rule Mining by Aprior method has been one of the popular data mining techniques for decades, where knowledge in the form of item-association rules is harvested from a dataset. The quality of item-association rules nevertheless depends on the concentration of frequent items from the input dataset. When the dataset becomes large, the items are scattered far apart. It is known from previous literature that clustering helps produce some data groups which are concentrated with frequent items. Among all the data clusters generated by a clustering algorithm, there must be one or more clusters which contain suitable and frequent items. In turn, the association rules that are mined from such clusters would be assured of better qualities in terms of high confidence than those mined from the whole dataset. However, it is not known in advance which cluster is the suitable one until all the clusters are tried by association rule mining. It is time consuming if they were to be tested by brute-force. In this paper, a statistical property called prior probability is investigated with respect to selecting the best out of many clusters by a clustering algorithm as a pre-processing step before association rule mining. Experiment results indicate that there is correlation between prior probability of the best cluster and the relatively high quality of association rules generated from that cluster. The results are significant as it is possible to know which cluster should be best used for association rule mining instead of testing them all out exhaustively.

It is plausible to assume that the component waves in ECG signals constitute a unique human characteristic because morphology and amplitudes of recorded beats are governed by multiple individual factors. According to the best of our knowledge, the issue of automatically classifying different ’identities’ of QRS morphology has not been explored within the literature. This work proposes five alternative mathematical models for representing different QRS morphologies providing the extraction of a set of features related to QRS shape. The technique incorporates mechanisms of combining the mathematical functions Gaussian, Mexican-Hat and Rayleigh probability density function and also a mechanism for clipping the waveform of those functions. The searching for the optimal parameters which minimize the normalized RMS error between each mathematical model and a given QRS search window enables to find an optimal model. Such modeling behaves as a robust alternative for delineating heartbeats, classifying beat morphologies, detecting subtle and anomalous changes, compression of QRS complex windows among others. The validation process evaluates the ability of each model to represent different QRS morphology classes within 159 full ECG signal records from QT database and 584 QRS search windows from MIT-BIH Arrhythmia database. From the experimental results, we rank the winning rates for which each mathematical model best models and also discriminates the most predominant QRS morphologies Rs, rS, RS, qR, qRs, R, rR’s and QS. Furthermore, the average time errors computed for QRS onset and offset locations when using the corresponding winner mathematical models for delineation purposes were, respectively, 12.87±8.5 ms and 1.47±10.06 ms.

<jats:title>Abstract</jats:title><jats:p>This research unveils to predict consumer ad preferences by detecting seven basic emotions, attention and engagement triggered by advertising through the analysis of two specific physiological monitoring tools, electrodermal activity (EDA), and Facial Expression Analysis (FEA), applied to video advertising, offering a twofold contribution of significant value. First, to identify the most relevant physiological features for consumer preference prediction. We integrated a statistical module encompassing inferential and exploratory analysis tools, which identified emotions such as Joy, Disgust, and Surprise, enabling the statistical differentiation of preferences concerning various advertisements. Second, we present an artificial intelligence (AI) system founded on machine learning techniques, encompassing k‐Nearest Neighbors, Support Vector Machine, and Random Forest (RF). Our findings show that the RF technique emerged as the top performer, boasting an 81% Accuracy, 84% Precision, 79% Recall, and an F1‐score of 81% in predicting consumer preferences. In addition, our research proposes an eXplainable AI module based on feature importance, which discerned Attention, Engagement, Joy, and Disgust as the four most pivotal features influencing consumer ad preference prediction. The results indicate that computerized intelligent systems based on EDA and FEA data can be used to predict consumer ad preferences based on videos and effectively used as supporting tools for marketing specialists.</jats:p>

<jats:p>Facial expression recognition (FER) is essential for discerning human emotions and is applied extensively in big data analytics, healthcare, security, and user experience enhancement. This study presents a comprehensive evaluation of ten state-of-the-art deep learning models—VGG16, VGG19, ResNet50, ResNet101, DenseNet, GoogLeNet V1, MobileNet V1, EfficientNet V2, ShuffleNet V2, and RepVGG—on the task of facial expression recognition using the FER2013 dataset. Key performance metrics, including test accuracy, training time, and weight file size, were analyzed to assess the learning efficiency, generalization capabilities, and architectural innovations of each model. EfficientNet V2 and ResNet50 emerged as top performers, achieving high accuracy and stable convergence using compound scaling and residual connections, enabling them to capture complex emotional features with minimal overfitting. DenseNet, GoogLeNet V1, and RepVGG also demonstrated strong performance, leveraging dense connectivity, inception modules, and re-parameterization techniques, though they exhibited slower initial convergence. In contrast, lightweight models such as MobileNet V1 and ShuffleNet V2, while excelling in computational efficiency, faced limitations in accuracy, particularly in challenging emotion categories like “fear” and “disgust”. The results highlight the critical trade-offs between computational efficiency and predictive accuracy, emphasizing the importance of selecting appropriate architecture based on application-specific requirements. This research contributes to ongoing advancements in deep learning, particularly in domains such as facial expression recognition, where capturing subtle and complex patterns is essential for high-performance outcomes.</jats:p>

The gold standard to detect SARS-CoV-2 infection consider testing methods based on Polymerase Chain Reaction (PCR). Still, the time necessary to confirm patient infection can be lengthy, and the process is expensive. On the other hand, X-Ray and CT scans play a vital role in the auxiliary diagnosis process. Hence, a trusted automated technique for identifying and quantifying the infected lung regions would be advantageous. Chest X-rays are two-dimensional images of the patient’s chest and provide lung morphological information and other characteristics, like ground-glass opacities (GGO), horizontal linear opacities, or consolidations, which are characteristics of pneumonia caused by COVID-19. But before the computerized diagnostic support system can classify a medical image, a segmentation task should usually be performed to identify relevant areas to be analyzed and reduce the risk of noise and misinterpretation caused by other structures eventually present in the images. This chapter presents an AI-based system for lung segmentation in X-ray images using a U-net CNN model. The system’s performance was evaluated using metrics such as cross-entropy, dice coefficient, and Mean IoU on unseen data. Our study divided the data into training and evaluation sets using an 80/20 train-test split method. The training set was used to train the model, and the evaluation test set was used to evaluate the performance of the trained model. The results of the evaluation showed that the model achieved a Dice Similarity Coefficient (DSC) of 95%, Cross entropy of 97%, and Mean IoU of 86%.

Consumers' selections and decision-making processes are some of the most exciting and challenging topics in neuromarketing, sales, and branding. From a global perspective, multicultural influences and societal conditions are crucial to consider. Neuroscience applications in international marketing and consumer behavior is an emergent and multidisciplinary field aiming to understand consumers' thoughts, reactions, and selection processes in branding and sales. This study focuses on real-time monitoring of different physiological signals using eye-tracking, facial expressions recognition, and Galvanic Skin Response (GSR) acquisition methods to analyze consumers' responses, detect emotional arousal, measure attention or relaxation levels, analyze perception, consciousness, memory, learning, motivation, preference, and decision-making. This research aimed to monitor human subjects' reactions to these signals during an experiment designed in three phases consisting of different branding advertisements. The nonadvertisement exposition was also monitored while gathering survey responses at the end of each phase. A feature extraction module with a data analytics module was implemented to calculate statistical metrics and decision-making supporting tools based on Principal Component Analysis (PCA) and Feature Importance (FI) determination based on the Random Forest technique. The results indicate that when compared to image ads, video ads are more effective in attracting consumers' attention and creating more emotional arousal.

The continuous development of robust machine learning algorithms in recent years has helped to improve the solutions of many studies in many fields of medicine, rapid diagnosis and detection of high-risk patients with poor prognosis as the coronavirus disease 2019 (COVID-19) spreads globally, and also early prevention of patients and optimization of medical resources. Here, we propose a fully automated machine learning system to classify the severity of COVID-19 from electrocardiogram (ECG) signals. We retrospectively collected 100 5-minute ECGs from 50 patients in two different positions, upright and supine. We processed the surface ECG to obtain QRS complexes and HRV indices for RR series, including a total of 43 features. We compared 19 machine learning classification algorithms that yielded different approaches explained in a methodology session.

In 2020, the World Health Organization declared the Coronavirus Disease 19 a global pandemic. While detecting COVID-19 is essential in controlling the disease, prognosis prediction is crucial in reducing disease complications and patient mortality. For that, standard protocols consider adopting medical imaging tools to analyze cases of pneumonia and complications. Nevertheless, some patients develop different symptoms and/or cannot be moved to a CT-Scan room. In other cases, the devices are not available. The adoption of ambulatory monitoring examinations, such as Electrocardiography (ECG), can be considered a viable tool to address the patient’s cardiovascular condition and to act as a predictor for future disease outcomes. In this investigation, ten non-linear features (Energy, Approximate Entropy, Logarithmic Entropy, Shannon Entropy, Hurst Exponent, Lyapunov Exponent, Higuchi Fractal Dimension, Katz Fractal Dimension, Correlation Dimension and Detrended Fluctuation Analysis) extracted from 2 ECG signals (collected from 2 different patient’s positions). Windows of 1 second segments in 6 ways of windowing signal analysis crops were evaluated employing statistical analysis. Three categories of outcomes are considered for the patient status: Low, Moderate, and Severe, and four combinations for classification scenarios are tested:  (Low vs. Moderate, Low vs. Severe, Moderate vs. Severe) and 1 Multi-class comparison  (All vs. All)). The results indicate that some statistically significant parameter distributions were found for all comparisons.  (Low vs. Moderate—Approximate Entropy p-value = 0.0067 < 0.05, Low vs. Severe—Correlation Dimension p-value = 0.0087 < 0.05, Moderate vs. Severe—Correlation Dimension p-value = 0.0029 < 0.05, All vs. All—Correlation Dimension p-value = 0.0185 < 0.05. The non-linear analysis of the time-frequency representation of the ECG signal can be considered a promising tool for describing and distinguishing the COVID-19 severity activity along its different stages.

Small and medium-sized enterprises (SMEs) can benefit significantly from open innovation by gaining access to a broader range of resources and expertise using absorptive capacitive, and increasing their visibility and reputation. Nevertheless, multiple barriers impact their capacity to absorb new technologies or adapt to develop them. This paper aims to perform an analysis of relevant topics and trends in Open Innovation (OI) and Absorptive Capacity (AC) in SMEs based on a bibliometric review identifying relevant authors and countries, and highlighting significant research themes and trends. The defined string query is submitted to the Web of Science database, and the bibliometric analysis using VOSviewer software. The results indicate that the number of scientific publications has consistently increased during the past decade, indicating a growing interest of the scientific community, reflecting the industry interest and possibly adoption of OI, considering Absorptive. This bibliometric analysis can provide insights on the most relevant regions the research areas are under intensive development.

<jats:p>Causal machine learning is an approach that combines causal inference and machine learning to understand and utilize causal relationships in data. In current research and applications, traditional machine learning and deep learning models always focus on prediction and pattern recognition. In contrast, causal machine learning goes a step further by revealing causal relationships between different variables. We explore a novel concept called Double Machine Learning that embraces causal machine learning in this research. The core goal is to select independent variables from a gesture identification problem that are causally related to final gesture results. This selection allows us to classify and analyze gestures more efficiently, thereby improving models’ performance and interpretability. Compared to commonly used feature selection methods such as Variance Threshold, Select From Model, Principal Component Analysis, Least Absolute Shrinkage and Selection Operator, Artificial Neural Network, and TabNet, Double Machine Learning methods focus more on causal relationships between variables rather than correlations. Our research shows that variables selected using the Double Machine Learning method perform well under different classification models, with final results significantly better than those of traditional methods. This novel Double Machine Learning-based approach offers researchers a valuable perspective for feature selection and model construction. It enhances the model’s ability to uncover causal relationships within complex data. Variables with causal significance can be more informative than those with only correlative significance, thus improving overall prediction performance and reliability.</jats:p>

Consumer neuroscience analyzes individuals’ preferences through the assessment of physiological data monitoring, considering brain activity or other bioinformation to assess purchase decisions. Traditional marketing tactics include customer surveys, product evaluations, and comments. For product or brand marketing and mass production, it is important to understand consumer neurological responses when seeing an ad or testing a product. In this work, we use the bi-clustering method to reduce EEG noise and automatic machine learning to classify brain responses. We analyze a neuromarketing EEG dataset that contains EEG data from product evaluations from 25 participants, collected with a 14 channel Emotiv Epoch + device, while examining consumer items. Four components comprised the research methodology. Initially, the Welch Transform was used to filter the EEG raw data. Second, the best converted signal biclusterings are used to train different classification models. Each biclustering is evaluated with a separate classifier, considering F1-Score. After that, the H2O.ai AutoML library is used to select the optimal biclustering and models. Instead of traditional procedures, two thresholds are used. First-threshold values indicate customer satisfaction. Low values of the second threshold reflect consumer dissatisfaction. Values between the first and second criteria are classified as uncertain values. We outperform the state of the art with a 0.95 F1-Score value.

<jats:p>Detecting emotions is a growing field aiming to comprehend and interpret human emotions from various data sources, including text, voice, and physiological signals. Electroencephalogram (EEG) is a unique and promising approach among these sources. EEG is a non-invasive monitoring technique that records the brain’s electrical activity through electrodes placed on the scalp’s surface. It is used in clinical and research contexts to explore how the human brain responds to emotions and cognitive stimuli. Recently, its use has gained interest in real-time emotion detection, offering a direct approach independent of facial expressions or voice. This is particularly useful in resource-limited scenarios, such as brain–computer interfaces supporting mental health. The objective of this work is to evaluate the classification of emotions (positive, negative, and neutral) in EEG signals using machine learning and deep learning, focusing on Graph Convolutional Neural Networks (GCNN), based on the analysis of critical attributes of the EEG signal (Differential Entropy (DE), Power Spectral Density (PSD), Differential Asymmetry (DASM), Rational Asymmetry (RASM), Asymmetry (ASM), Differential Causality (DCAU)). The electroencephalography dataset used in the research was the public SEED dataset (SJTU Emotion EEG Dataset), obtained through auditory and visual stimuli in segments from Chinese emotional movies. The experiment employed to evaluate the model results was “subject-dependent”. In this method, the Deep Neural Network (DNN) achieved an accuracy of 86.08%, surpassing SVM, albeit with significant processing time due to the optimization characteristics inherent to the algorithm. The GCNN algorithm achieved an average accuracy of 89.97% in the subject-dependent experiment. This work contributes to emotion detection in EEG, emphasizing the effectiveness of different models and underscoring the importance of selecting appropriate features and the ethical use of these technologies in practical applications. The GCNN emerges as the most promising methodology for future research.</jats:p>

The scientific literature indicates that pregnant women with COVID-19 are at an increased risk for developing more severe illness conditions when compared with non-pregnant women. The risk of admission to an ICU (Intensive Care Unit) and the need for mechanical ventilator support is three times higher. More significantly, statistics indicate that these patients are also at 70% increased risk of evolving to severe states or even death. In addition, other previous illnesses and age greater than 35 years old increase the risk for the mother and the fetus, including a higher number of cesarean sections, higher systolic and diastolic maternal blood pressure, increasing the risk of eclampsia, and, in some cases, preterm birth. Additionally, pregnant women have more Emotional lability/fluctuations (between positive and negative feelings) during the entire pregnancy. The emotional instability and brain fog that takes place during gestation may open vulnerability for neuropsychiatric symptoms of long COVID, which this population was not studied in depth. The present Chapter characterizes the database presented in this work with clinical and survey data collected about emotions and feelings using the Coronavirus Perinatal Experiences—Impact Survey (COPE-IS). Pregnant women with or without COVID-19 symptoms who gave birth at the Assis Chateaubriand Maternity Hospital (MEAC), a public maternity of the Federal University of Ceara, Brazil, were recruited. In total, 72 mother-infant dyads were included in the study and are considered in this exploratory analysis. The participants have undergone serological tests for SARS-CoV-2 antibody detection and a nasopharyngeal swab test for COVID-19 diagnoses by RT-PCR. A comprehensive Exploratory Data Analysis (EDA) is performed using frequency distribution analysis of multiple types of variables generated from numerical data, multiple-choice, categorized, and Likert-scale questions.

This work provides a comprehensive systematic review of optimization techniques using artificial intelligence (AI) for energy storage systems within renewable energy setups. The primary goals are to evaluate the latest technologies employed in forecasting models for renewable energy generation, load forecasting, and energy storage systems, alongside their construction parameters and optimization methods. The review highlights the progress achieved, identifies current challenges, and explores future research directions. Despite the extensive application of machine learning (ML) and deep learning (DL) in renewable energy generation, consumption patterns, and storage optimization, few studies integrate these three aspects simultaneously, underscoring the significance of this work. The review encompasses studies from Web of Science, Scopus, and Science Direct up to December 2023, including works scheduled for publication in 2024. Each study related to renewable energy storage was individually analyzed to assess its objectives, methodology, and results. The findings reveal useful insights for developing AI models aimed at optimizing storage systems. However, critical areas need further exploration, such as real-time forecasting, long-term storage predictions, hybrid neural networks for demand-based generation forecasting, and the evaluation of various storage scales and battery technologies. The review also notes a significant gap in research on large-scale storage systems in Brazil and Latin America. In conclusion, the study emphasizes the need for continued research and the development of new algorithms to address existing limitations in the field.

Introduction: SARS-CoV-2, a virus responsible for the emergence of the life-threatening disease known as COVID-19, exhibits a diverse range of clinical manifestations. The spectrum of symptoms varies widely, encompassing mild to severe presentations, while a considerable portion of the population remains asymptomatic. COVID-19, primarily a respiratory virus, has been linked to cardiovascular complications in some patients. Notably, cardiac issues can also arise after recovery, contributing to post-acute COVID-19 syndrome, a significant concern for patient health. The present study intends to evaluate the post-acute COVID-19 syndrome cardiovascular effect through ECG by comparing patients affected with cardiac diseases without COVID-19 diagnosis report (class 1) and patients with cardiac pathologies who present post-acute COVID-19 syndrome (class 2). Methods: From 2 body positions, a total of 10 non-linear features, extracted every 1 second under a multi-band analysis performed by Discrete Wavelet Transform (DWT), have been compressed by 6 statistical metrics to serve as inputs for an individual feature analysis by the means of Mann-Whitney U-test and XROC classification. Results and Discussion: 480 Mann-Whitney U-test statistical analyses and XROC discrimination approaches have been done. The percentage of statistical analysis with significant differences (p<0.05) was 30.42% (146 out of 480). The best overall results were obtained by approximating the feature Energy, with the data compressor Kurtosis in the body position Down. Those results were 83.33% of Accuracy, 83.33% of Sensitivity, 83.33% of Specificity and 87.50% of AUC. Conclusions: The results show that the applied methodology can be a way to show changes in cardiac behaviour provoked by post-acute COVID-19 syndrome.