Machine learning (ML), more recently, deep learning, in combination with the two mainstream approaches of technical and fundamental analysis, has been widely applied to stock price prediction. Since technical analysis remains the dominant paradigm in this literature, comprehensive surveys centered exclusively on fundamental indicators are notably absent. This study addresses this gap through a systematic review of 54 empirical studies published between 2014 and 2025, identified from an initial pool of 2263 papers indexed in Scopus and Web of Science. The review catalogs 544 fundamental indicators across ten categories and examines their alignment with seven model families, alongside feature selection techniques, evaluation criteria, and data sources. Valuation metrics and profitability measures dominate the literature, with the P/E ratio and ROA being the most frequently employed indicators across price, return, and signal prediction tasks. Among model architectures, ANNs and Random Forest lead traditional ML approaches, while LSTM networks dominate deep learning applications. Using chi-square tests of independence, point-biserial correlations, and adjusted standardized residuals, we find a significant association between model category and prediction task, LSTM networks show a strong affinity for price prediction, Random Forests for return prediction, and boosting methods a negative association with price prediction. Transformer-based architectures show promise but have yet to establish dominance, and LLMs currently function more effectively in auxiliary roles such as financial statement interpretation and sentiment extraction than as standalone predictive engines. Accuracy, RMSE, and the Sharpe Ratio serve as the standard evaluation metrics across the reviewed literature. A sensitivity analysis across three inclusion scenarios confirms the robustness of core findings. This survey provides a unified framework for model-indicator alignment and offers practical guidance for feature selection, evaluation standards, and future research priorities.

Schwannomatosis is a rare genetic disorder characterized by the development of multiple, painful schwannomas, presenting with high variability in clinical manifestation and management complexity. To support clinical decision-making, we previously developed an Explainable Artificial Intelligence (XAI) system based on the COGNICA framework of cognitive argumentation. The system integrates the European Reference Network (ERN) GENTURIS clinical guidelines with structured, interpretable reasoning processes, enabling transparent and traceable medical recommendations. While prior work established the system’s technical feasibility, its usability and clinical acceptability among domain experts had not been systematically assessed. This study presents the results of an expert-based evaluation using a structured questionnaire that combines standardized instruments—Visual Analogue Scales (VAS), System Usability Scale (SUS), Questionnaire for User Interface Satisfaction (QUIS), User Experience Questionnaire (UEQ), and Treatment Acceptability Rating Form (TARF)—alongside open-ended feedback. Four experts from neurology, and genetics participated in the evaluation. Results demonstrate strong satisfaction and comprehension, confirming that explainable AI tools like this system can enhance medical reasoning in rare diseases. However, their effective integration into clinical workflows will require iterative refinement, clinician training, and validation in real-world settings.

Over the past decade, volatility modeling has gained increasing importance in quantitative finance, significantly influencing risk management, investment strategies, and policymaking. Traditionally, academic research has emphasized linear models for forecasting realized volatility. However, advances in computational power have enabled the development of more sophisticated approaches, including machine and deep learning models. Despite these advancements, no comprehensive survey currently compares traditional linear methods with newer models in the context of realized volatility forecasting. This survey addresses that gap by analyzing all models used to forecast realized volatility in academic literature from 2000 to the first half of 2024. It highlights key academic contributions and examines the empirical characteristics of realized volatility. Our findings indicate that the top-performing hybrid convolutional neural network and long short-term memory model surpasses other models in forecasting accuracy. These results highlight the ability of such models to capture a broader range of factors driving realized volatility, affirming their valuable role in quantitative modeling.

Pain assessment is a critical aspect of medical practice, directly influencing patient treatment and quality of life. Traditional pain evaluation methods, such as the Numerical Rating Scale (NRS), Visual Analog Scale (VAS), and Verbal Rating Scale (VRS), are subjective and often unreliable, especially for non-verbal, unconscious, or cognitively impaired patients. This study introduces an objective pain measurement model using advanced machine learning techniques, specifically, Convolutional Neural Networks (CNNs), and Vision Transformers (ViTs), to analyze facial expressions. We compared the performance of CNNs, VGG16, Convolutional Vision Transformer (CvT), and MobileViT, in classifying pain intensity based on facial images captured during peak pain and no-pain moments. The models were trained and evaluated on the BioVid Heat Pain Database, which comprises facial recordings of 87 participants experiencing varying pain intensities. The comparative analysis of the CNN, VGG16, CvT, and MobileViT50 models reveal distinct differences in their performance metrics. Among the four models, the CNN model achieved the highest average accuracy at 0.71, demonstrating better performance in correctly classifying both pain and no pain images. The CvT model followed closely with an average accuracy of 0.69, indicating that it also performed well, although slightly less effectively than CNN. MobileViT50, with an accuracy of 0.60, and VGG16, with 0.56, performed significantly lower, suggesting that these models struggled more with accurately classifying the data. These results highlight the potential of automated pain assessment technologies to provide consistent and objective evaluations, which can be particularly beneficial in clinical environments for non-communicative patients. Future research will explore the integration of multimodal data to further enhance the robustness of pain detection systems.

Effective pain assessment is crucial in clinic practice, as it directly impacts treatment decisions and patients' outcomes. Regular pain evaluation relies on self-reported scales, which are often limited by subjectivity and variability. This study investigates advanced machine learning techniques, specifically 3DCNN, TSN, ConvLSTM and VideoMAE architecture, for automated pain expression intensity assessment from videos. Using videos capturing peak pain and pain-free states, we trained and evaluated these models on the Biovid Heat Pain Database, with distinct participant sets for each phase. VideoMAE model demonstrated promising classification performance, achieving an average accuracy, F1 score, recall, and precision of 0.71 in distinguishing between maximum pain and no pain from videos. Among the models tested, four classification architectures demonstrated better performance with regards to accuracy compared to previous studies, underscoring their potential advantage in clinical pain assessment applications. Further work is necessary to expand dataset volume and diversity to improve model generalization and minimize biases. Additionally, exploring emerging lightweight, real-time models is crucial to enable seamless integration into mobile and wearable systems.

Visual place recognition is a critical task in computer vision, especially for localization and navigation systems. Existing methods often rely on contrastive learning: image descriptors are trained to have small distance for similar images and larger distance for dissimilar ones in a latent space. However, this approach struggles to ensure accurate distance-based image similarity representation, particularly when training with binary pairwise labels, and complex re-ranking strategies are required. This work introduces a fresh perspective by framing place recognition as a regression problem, using camera field-of-view overlap as similarity ground truth for learning. By optimizing image descriptors to align directly with graded similarity labels, this approach enhances ranking capabilities without expensive re-ranking, offering data-efficient training and strong generalization across several benchmark datasets.

In recent years, brain research has indisputably entered a new epoch, driven by substantial methodological advances and digitally enabled data integration and modelling at multiple scales—from molecules to the whole brain. Major advances are emerging at the intersection of neuroscience with technology and computing. This new science of the brain combines high-quality research, data integration across multiple scales, a new culture of multidisciplinary large-scale collaboration, and translation into applications. As pioneered in Europe’s Human Brain Project (HBP), a systematic approach will be essential for meeting the coming decade’s pressing medical and technological challenges. The aims of this paper are to: develop a concept for the coming decade of digital brain research, discuss this new concept with the research community at large, identify points of convergence, and derive therefrom scientific common goals; provide a scientific framework for the current and future development of EBRAINS, a research infrastructure resulting from the HBP’s work; inform and engage stakeholders, funding organisations and research institutions regarding future digital brain research; identify and address the transformational potential of comprehensive brain models for artificial intelligence, including machine learning and deep learning; outline a collaborative approach that integrates reflection, dialogues, and societal engagement on ethical and societal opportunities and challenges as part of future neuroscience research.

AbstractForecasting changes in stock prices is extremely challenging given that numerous factors cause these prices to fluctuate. The random walk hypothesis and efficient market hypothesis essentially state that it is not possible to systematically, reliably predict future stock prices or forecast changes in the stock market overall. Nonetheless, machine learning (ML) techniques that use historical data have been applied to make such predictions. Previous studies focused on a small number of stocks and claimed success with limited statistical confidence. In this study, we construct feature vectors composed of multiple previous relative returns and apply the random forest (RF), support vector machine (SVM), and long short-term memory (LSTM) ML methods as classifiers to predict whether a stock can return 2% more than its index in the following 10 days. We apply this approach to all S&P 500 companies for the period 2017–2022. We assess performance using accuracy, precision, and recall and compare our results with a random choice strategy. We observe that the LSTM classifier outperforms RF and SVM, and the data-driven ML methods outperform the random choice classifier (p = 8.46e−17 for accuracy of LSTM). Thus, we demonstrate that the probability that the random walk and efficient market hypotheses hold in the considered context is negligibly small.

Visual place recognition (VPR) is a fundamental task of computer vision for visual localization. Existing methods are trained using image pairs that either depict the same place or not. Such a binary indication does not consider continuous relations of similarity between images of the same place taken from different positions, determined by the continuous nature of camera pose. The binary similarity induces a noisy supervision signal into the training of VPR methods, which stall in local minima and require expensive hard mining algorithms to guarantee convergence. Motivated by the fact that two images of the same place only partially share visual cues due to camera pose differences, we deploy an automatic re-annotation strategy to re-label VPR datasets. We compute graded similarity labels for image pairs based on available localization metadata. Furthermore, we propose a new Generalized Contrastive Loss (GCL) that uses graded similarity labels for training contrastive networks. We demonstrate that the use of the new labels and GCL allow to dispense from hard-pair mining, and to train image descriptors that perform better in VPR by nearest neighbor search, obtaining superior or comparable results than methods that require expensive hard-pair mining and re-ranking techniques.

Various effects show that the visual perception of an edge or line can be influenced by other such stimuli in the surroundings. Such effects can be related to nonclassical receptive field (non-CRF) inhibition, also called surround suppression, that is found in most of the orientation selective neurones in the primary visual cortex. A mathematical model of non-CRF inhibition is presented. Non-CRF inhibition acts as a feature contrast computation for oriented stimuli: the response to an edge at a given position is suppressed by other edges in the surround. Consequently, it strongly reduces the responses to texture edges while scarcely affecting the responses to isolated contours. The biological utility of this neural mechanism might thus be that of improving contour (vs. texture) detection. The results of computer simulations based on the proposed model explain perceptual effects, such as orientation contrast pop-out, ‘social conformity’ of lines embedded in gratings, reduced saliency of contours surrounded by textures and decreased visibility of letters embedded in band-limited noise. The insights into the biological role of non-CRF inhibition can be utilised in machine vision. The proposed model is employed in a contour detection algorithm. Applied on natural images it outperforms previously known such algorithms in computer vision.

In stock market forecasting, the identification of critical features that affect the performance of machine learning (ML) models is crucial to achieve accurate stock price predictions. Several review papers in the literature have focused on various ML, statistical, and deep learning-based methods used in stock market forecasting. However, no survey study has explored feature selection and extraction techniques for stock market forecasting. This survey presents a detailed analysis of 32 research works that use a combination of feature study and ML approaches in various stock market applications. We conduct a systematic search for articles in the Scopus and Web of Science databases for the years 2011–2022. We review a variety of feature selection and feature extraction approaches that have been successfully applied in the stock market analyses presented in the articles. We also describe the combination of feature analysis techniques and ML methods and evaluate their performance. Moreover, we present other survey articles, stock market input and output data, and analyses based on various factors. We find that correlation criteria, random forest, principal component analysis, and autoencoder are the most widely used feature selection and extraction techniques with the best prediction accuracy for various stock market applications.

This paper proposes LTC-Mapping, a method for building object-oriented semantic maps that remain consistent in the long-term operation of mobile robots. Among the different challenges that compromise this aim, LTC-Mapping focuses on two of the more relevant ones: preventing duplicate instances of objects (instance duplication) and handling dynamic scenes. The former refers to creating multiple instances of the same physical object in the map, usually as a consequence of partial views or occlusions. The latter deals with the typical assumption made by object-oriented mapping methods that the world is static, resulting in outdated representations when the objects change their positions. To face these issues, we model the detected objects with 3D bounding boxes, and analyze the visibility of their vertices to detect occlusions and partial views. Besides this geometric modeling, the boxes are augmented with semantic information regarding the categories of the objects they represent. Both the geometric entities (bounding boxes) and their semantic content are propagated over time through data association and a fusion technique. In addition, in order to keep the map curated, the non-detection of objects in the areas where they should appear is also considered, proposing a mechanism that removes them from the map once there is evidence that they have been moved (i.e., multiple non-detections occur). To validate our proposal, a number of experiments have been carried out using the Robot@VirtualHome ecosystem, comparing its performance with a state-of-the-art alternative. The results report a superior performance of LTC-Mapping when modeling both geometric and semantic information of objects, and also support its online execution.

As dimensions of datasets in predictive modelling continue to grow, feature selection becomes increasingly practical. Datasets with complex feature interactions and high levels of redundancy still present a challenge to existing feature selection methods. We propose a novel framework for feature selection that relies on boosting, or sample re-weighting, to select sets of informative features in classification problems. The method uses as its basis the feature rankings derived from fast and scalable tree-boosting models, such as XGBoost. We compare the proposed method to standard feature selection algorithms on 9 benchmark datasets. We show that the proposed approach reaches higher accuracies with fewer features on most of the tested datasets, and that the selected features have lower redundancy.

Simulations and synthetic datasets have historically empower the research in different service robotics-related problems, being revamped nowadays with the utilization of rich virtual environments. However, with their use, special attention must be paid so the resulting algorithms are not biased by the synthetic data and can generalize to real world conditions. These aspects are usually compromised when the virtual environments are manually designed. This article presents Robot@VirtualHome, an ecosystem of virtual environments and tools that allows for the management of realistic virtual environments where robotic simulations can be performed. Here “realistic” means that those environments have been designed by mimicking the rooms’ layout and objects appearing in 30 real houses, hence not being influenced by the designer’s knowledge. The provided virtual environments are highly customizable (lighting conditions, textures, objects’ models, etc.), accommodate meta-information about the elements appearing therein (objects’ types, room categories and layouts, etc.), and support the inclusion of virtual service robots and sensors. To illustrate the possibilities of Robot@VirtualHome we show how it has been used to collect a synthetic dataset, and also exemplify how to exploit it to successfully face two service robotics-related problems: semantic mapping and appearance-based localization.

Dystocia or difficult calving in cattle is detrimental to the health of the afflicted cows and has a negative economic impact on the dairy industry. The goal of this study was to create a data-driven tool for predicting the calving difficulty of non-heifer cows using input variables that are known prior to the moment of insemination. Compared to past studies, we excluded input variables that can only be known during or after insemination, such as birth weight and gestation length. This makes the model suitable for informing mating decisions that could reduce the incidence of difficult calvings or mitigate their consequences. We used a dataset consisting of 131,527 calving records of Holstein cattle, from which we derived a total of 274 phenotypic features and estimated breeding values. The distribution of classes in the dataset was 96.7 % normal calvings, and 3.3 % difficult calvings. We used a gradient boosted trees (XGBoost) as the learning model and a bagging ensemble approach to deal with the extreme class imbalance. The model achieved an average area under the ROC curve of 0.73 on unseen test data. Using feature importance analysis, we identified a number of features that have a high discriminatory value for calving difficulty, including maternal and paternal breeding values, and past phenotypic measurements of the cow.

This paper proposes a method to enhance video object detection for indoor environments in robotics. Concretely, it exploits knowledge about the camera motion between frames to propagate previously detected objects to successive frames. The proposal is rooted in the concepts of planar homography to propose regions of interest where to find objects, and recursive Bayesian filtering to integrate observations over time. The proposal is evaluated on six virtual, indoor environments, accounting for the detection of nine object classes over a total of \(\sim \)7k frames. Results show that our proposal improves the recall and the F1-score by a factor of 1.41 and 1.27, respectively, as well as it achieves a significant reduction of the object categorization entropy (58.8%) when compared to a two-stage video object detection method used as baseline, at the cost of small time overheads (120 ms) and precision loss (0.92).

The u-serrated immunodeposition pattern in direct immunofluorescence (DIF) microscopy is a recognizable feature and confirmative for the diagnosis of epidermolysis bullosa acquisita (EBA). Due to unfamiliarity with serrated patterns, serration pattern recognition is still of limited use in routine DIF microscopy. The objective of this study was to investigate the feasibility of using convolutional neural networks (CNNs) for the recognition of u-serrated patterns that can assist in the diagnosis of EBA. The nine most commonly used CNNs were trained and validated by using 220,800 manually delineated DIF image patches from 106 images of 46 different patients. The data set was split into 10 subsets: nine training subsets from 42 patients to train CNNs and the last subset from the remaining four patients for a validation data set of diagnostic accuracy. This process was repeated 10 times with a different subset used for validation. The best-performing CNN achieved a specificity of 89.3% and a corresponding sensitivity of 89.3% in the classification of u-serrated DIF image patches, an expert level of diagnostic accuracy. Experiments and results show the effectiveness of CNN approaches for u-serrated pattern recognition with a high accuracy. The proposed approach can assist clinicians and pathologists in recognition of u-serrated patterns in DIF images and facilitate the diagnosis of EBA.

Efficient yet accurate extraction of depth from stereo image pairs is required by systems with low power resources, such as robotics and embedded systems. State-of-the-art stereo matching methods based on convolutional neural networks require intensive computations on GPUs and are difficult to deploy on embedded systems. In this paper, we propose MTStereo2.0, an improved version of the MTStereo stereo matching method, which includes a more robust context-driven cost function, better detection of incorrect matches and the computation of disparity at pixel level. MTStereo provides accurate sparse and semi-dense depth estimation and does not require intensive GPU computations. We tested it on several benchmark data sets, namely KITTI 2015, Driving, FlyingThings3D, Middlebury 2014, Monkaa and the TrimBot2020 garden data sets, and achieved competitive accuracy. The code is available at https://github.com/rbrandt1/MaxTreeS.

Semantic maps augment traditional representations of robot workspaces, typically based on their geometry and/or topology, with meta-information about the properties, relations and functionalities of their composing elements. A piece of such information could be: fridges are appliances typically found in kitchens and employed to keep food in good condition. Thereby, semantic maps allow for the execution of high-level robotic tasks in an efficient way, e.g. “Hey robot, Store the leftover salad”. This paper presents ViMantic, a novel semantic mapping architecture for the building and maintenance of such maps, which brings together a number of features as demanded by modern mobile robotic systems, including: (i) a formal model, based on ontologies, which defines the semantics of the problem at hand and establishes mechanisms for its manipulation; (ii) techniques for processing sensory information and automatically populating maps with, for example, objects detected by cutting-edge CNNs; (iii) distributed execution capabilities through a client–server design, making the knowledge in the maps accessible and extendable to other robots/agents; (iv) a user interface that allows for the visualization and interaction with relevant parts of the maps through a virtual environment; (v) public availability, hence being ready to use in robotic platforms. The suitability of ViMantic has been assessed using Robot@Home, a vast repository of data collected by a robot in different houses. The experiments carried out consider different scenarios with one or multiple robots, from where we have extracted satisfactory results regarding automatic population, execution times, and required size in memory of the resultant semantic maps.