Search Results (121)

Road traffic accident severity prediction is crucial for implementing effective safety measures and proactive traffic management strategies. Existing methods often treat this as a nominal classification problem and use traditional feature selection techniques. However, ordinal classification methods that account for the ordered nature of accident severity (e.g., slight < serious < fatal injuries) in feature selection still need to be investigated thoroughly. In this study, we propose a novel approach, the Ordinal Random Tree with Rank-Oriented Feature Selection (ORT-ROFS), which utilizes the inherent ordering of class labels both in the feature selection and prediction stages for accident severity classification. The proposed approach enhances the model performance by separately determining feature importance based on severity levels. The experiments demonstrated the effectiveness of ORT-ROFS with an accuracy of 87.19%. According to the results, the proposed method improved prediction accuracy by 10.81% over state-of-the-art studies on average on different train–test split ratios. In addition, it achieved an average improvement of 4.58% in accuracy over traditional methods. These findings suggest that ORT-ROFS is a promising approach for accurate accident severity prediction, supporting road safety planning and intervention strategies. Full article

Recreational Ecological Parks (REPs) play a pivotal role in advancing sustainable development by promoting ecotourism, conserving biodiversity, and providing inclusive recreational opportunities with minimal environmental impact. Assessing the potential for REP development is crucial to optimize the sustainable use of natural resources and enhance ecological and socio-economic benefits. This study introduces a sustainability-oriented Spatial Multi-Criteria Decision Analysis (SMCDA) framework to evaluate and optimize REP development in Mazandaran Province, Iran. The framework incorporates 33 criteria across five categories: 13 for network connectivity and accessibility; 10 for facilities, services, and tourism potential; 4 for landform and land use; 3 for natural hazards; and 3 for climate conditions. Criteria were standardized using the minimum–maximum method and weighted based on expert input via the Best–Worst Method. A weighted linear combination approach was then applied to generate REP suitability maps. Existing recreational and tourism (R&T) sites were assessed against these maps, and a non-parametric bootstrapping method quantified uncertainties in suitability classifications. The results revealed that approximately 8%, 17%, 26%, 30%, and 19% of the study area fell into very low, low, moderate, high, and very high suitability classes, respectively. Among 151 R&T sites, 33% and 34% were situated in areas classified as very high and high suitability. Prediction rates were most accurate in the very high suitability category, while uncertainty analysis indicated a mean of 13% and a standard deviation of 1.7%, with uncertainties predominantly concentrated in lower suitability classes. The findings underscore the SMCDA framework’s efficacy in guiding sustainable REP development by identifying optimal sites and managing uncertainties. This study contributes to sustainability by integrating ecological, economic, and social dimensions into decision-making processes, thereby fostering resilience and long-term environmental stewardship in tourism planning. Full article

Compositional and anatomical studies of silicified wood have been carried out extensively all around the world. The classification of silicified wood as such deals with all the forms and phases of silica that come under its umbrella. One such class of silicified wood is fossil wood with a high content of quartz, and there are very limited mentions of this category of fossilized wood. The examined wood belongs to gymnosperm and comes from the Upper Triassic deposits of Madagascar. A fresh approach to such samples is adopted by studying the crystallographic texture of the fossil wood to understand the orientation of the crystals replacing the organic matter within the sample. This work focuses on crystallographic texture analysis based on pole figures measured by X-ray diffraction. The intensity of the pole density maxima on the pole figures measured on the heartwood surface part of the analyzed samples is higher than that on the sapwood. This affirms that the crystallographic texture is sharper at the heartwood part compared to the sapwood. The X-ray tomography study, conducted to understand the difference in mineral distribution within the sample, reveals a greater X-ray absorbing phase on the sapwood of both samples. This is due to the concentration of iron compounds, which both replace the remaining conductive structures of the wood and fill the cavities inside them. We believe that this research on silicified wood is the first research work that encompasses crystallographic texture analysis with pole figures, an approach not previously undertaken in similar studies. We hope that our research can be useful in understanding the processes of replacement of organic matter by minerals. Full article

This paper introduces a novel deep learning model for ECG signal classification using feature fusion. The proposed methodology transforms the ECG time series into a spectrogram image using a short-time Fourier transform (STFT). This spectrogram is further processed to generate a histogram of oriented gradients (HOG) and local binary pattern (LBP) features. Three separate 2D convolutional neural networks (CNNs) then analyze these three image representations in parallel. To enhance performance, the extracted features are concatenated before feeding them into a gated recurrent unit (GRU) model. The proposed approach is extensively evaluated on two ECG datasets (MIT-BIH + BIDMC and MIT-BIH) with three and five classes, respectively. The experimental results demonstrate that the proposed approach achieves superior classification accuracy compared to existing algorithms in the literature. This suggests that the model has the potential to be a valuable tool for accurate ECG signal classification, aiding in the diagnosis and treatment of various cardiovascular disorders. Full article

Automated assessment of surgical skills is crucial for the successful training of junior surgeons. Twenty-three medical students followed a structured training curriculum on a laparoscopic virtual reality (VR) simulator. Three surgical tasks with significant educational merit were considered (Tasks 5, 6, and 7). We evaluated seven machine learning (ML) models for classifying the students’ trials into two and three classes based on the progress of training (Beginning vs. End and Beginning vs. Middle vs. End). Additionally, we evaluated the same ML framework and a deep learning approach (LSTM) for predicting the remaining number of trials required to complete the training proficiently. A model-agnostic technique from the domain of explainable artificial intelligence (XAI) was also utilized to obtain interpretations of the employed black-box ML classifiers. For 2-class classification, the best model showed an accuracy of 97.1%, 96.9%, and 75.7% for Task 5, 6, and 7, respectively, whereas for 3-class classification, the corresponding accuracy was 96.3%, 95.9%, and 99.7%, respectively. The best regression algorithm was LSTM with a Mean Absolute Error of 4 (Task 5) and 3.6 trials (Tasks 6, 7). According to XAI, the kinematic parameters have a stronger impact on the classification decision than the goal-oriented metrics. Full article

One key aspect of the human experience is our ongoing stream of thoughts. These thoughts can be broadly categorized into various dimensions, which are associated with different impacts on mood, well-being, and productivity. While the past literature has often identified eye movements associated with a specific thought dimension (task-relatedness) during experimental tasks, few studies have determined if these various thought dimensions can be classified by oculomotor activity during naturalistic tasks. Employing thought sampling, eye tracking, and machine learning, we assessed the classification of nine thought dimensions (task-relatedness, freely moving, stickiness, goal-directedness, internal–external orientation, self-orientation, others orientation, visual modality, and auditory modality) across seven multi-day recordings of seven participants during self-selected computer tasks. Our analyses were based on a total of 1715 thought probes across 63 h of recordings. Automated binary-class classification of the thought dimensions was based on statistical features extracted from eye movement measures, including fixation and saccades. These features all served as input into a random forest (RF) classifier, which was then improved with particle swarm optimization (PSO)-based selection of the best subset of features for classifier performance. The mean Matthews correlation coefficient (MCC) values from the PSO-based RF classifier across the thought dimensions ranged from 0.25 to 0.54, indicating above-chance level performance in all nine thought dimensions across participants and improved performance compared to the RF classifier without feature selection. Our findings highlight the potential of machine learning approaches combined with eye movement measures for the real-time prediction of naturalistic ongoing thoughts, particularly in ecologically valid contexts. Full article

Putting sensors on the bodies of animals to automate animal activity recognition and gain insight into their behaviors can help improve their living conditions. Although previous hard-coded algorithms failed to classify complex time series obtained from accelerometer data, recent advances in deep learning have improved the task of animal activity recognition for the better. However, a comparative analysis of the generalizing capabilities of various models in combination with different input types has yet to be addressed. This study experimented with two techniques for transforming the segmented accelerometer data to make them more orientation-independent. The methods included calculating the magnitude of the three-axis accelerometer vector and calculating the Discrete Fourier Transform for both sets of three-axis data as the vector magnitude. Three different deep learning models were trained on this data: a Multilayer Perceptron, a Convolutional Neural Network, and an ensemble merging both called a hybrid Convolutional Neural Network. Besides mixed cross-validation, every model and input type combination was assessed on a goat-wise leave-one-out cross-validation set to evaluate its generalizing capability. Using orientation-independent data transformations gave promising results. A hybrid Convolutional Neural Network with $L^2$-norm as the input combined the higher classification accuracy of a Convolutional Neural Network with the lower standard deviation of a Multilayer Perceptron. Most of the misclassifications occurred for behaviors that display similar accelerometer traces and minority classes, which could be improved in future work by assembling larger and more balanced datasets. Full article

The epidemiological burden of liver steatosis associated with metabolic diseases is continuously growing worldwide and in all age classes. This condition generates possible progression of liver damage (i.e., inflammation, fibrosis, cirrhosis, hepatocellular carcinoma) but also independently increases the risk of cardio-metabolic diseases and cancer. In recent years, the terminological evolution from “nonalcoholic fatty liver disease” (NAFLD) to “metabolic dysfunction-associated fatty liver disease” (MAFLD) and, finally, “metabolic dysfunction-associated steatotic liver disease” (MASLD) has been paralleled by increased knowledge of mechanisms linking local (i.e., hepatic) and systemic pathogenic pathways. As a consequence, the need for an appropriate classification of individual phenotypes has been oriented to the investigation of innovative therapeutic tools. Besides the well-known role for lifestyle change, a number of pharmacological approaches have been explored, ranging from antidiabetic drugs to agonists acting on the gut–liver axis and at a systemic level (mainly farnesoid X receptor (FXR) agonists, PPAR agonists, thyroid hormone receptor agonists), anti-fibrotic and anti-inflammatory agents. The intrinsically complex pathophysiological history of MASLD makes the selection of a single effective treatment a major challenge, so far. In this evolving scenario, the cooperation between different stakeholders (including subjects at risk, health professionals, and pharmaceutical industries) could significantly improve the management of disease and the implementation of primary and secondary prevention measures. The high healthcare burden associated with MASLD makes the search for new, effective, and safe drugs a major pressing need, together with an accurate characterization of individual phenotypes. Recent and promising advances indicate that we may soon enter the era of precise and personalized therapy for MASLD/MASH. Full article

Diabetes is caused by insulin deficiency or impaired biological action, and long-term hyperglycemia leads to a variety of tissue damage and dysfunction. Therefore, the early prediction of diabetes and timely intervention and treatment are crucial. This paper proposes a robust framework for the prediction and diagnosis of type 2 diabetes (T2DM) to aid in diabetes applications in clinical diagnosis. The data-preprocessing stage includes steps such as outlier removal, missing value filling, data standardization, and assigning class weights to ensure the quality and consistency of the data, thereby improving the performance and stability of the model. This experiment used the National Health and Nutrition Examination Survey (NHANES) dataset and the publicly available PIMA Indian dataset (PID). For T2DM classification, we designed a convolutional neural network (CNN) and proposed a novel attention-oriented convolutional neural network (SECNN) through the channel attention mechanism. To optimize the hyperparameters of the model, we used grid search and K-fold cross-validation methods. In addition, we also comparatively analyzed various machine learning (ML) models such as support vector machine (SVM), logistic regression (LR), decision tree (DT), random forest (RF), and artificial neural network (ANN). Finally, we evaluated the performance of the model using performance evaluation metrics such as precision, recall, F1-Score, accuracy, and AUC. Experimental results show that the SECNN model has an accuracy of 94.12% on the NHANES dataset and an accuracy of 89.47% on the PIMA Indian dataset. SECNN models and CNN models show significant improvements in diabetes prediction performance compared to traditional ML models. The comparative analysis of the SECNN model and the CNN model has significantly improved performance, further verifying the advantages of introducing the channel attention mechanism. The robust diabetes prediction framework proposed in this article establishes an effective foundation for diabetes diagnosis and prediction, and has a positive impact on the development of health management and medical industries. Full article

Oriented object detection (OOD) can precisely detect objects with arbitrary direction in remote sensing images (RSIs). Up to now, the two-stage OOD methods have attracted more attention because of their high detection accuracy. However, the two-stage methods only rely on the features of each proposal for object recognition, which leads to the misclassification problem because of the intra-class diversity, inter-class similarity and clutter backgrounds in RSIs. To address the above problem, an OOD model combining scene classification is proposed. Considering the fact that each foreground object has a strong contextual relationship with the scene of the RSI, a scene classification branch is added to the baseline OOD model, and the scene classification result of input RSI is used to exclude the impossible categories. To focus on the hard instances and enhance the consistency between classification and regression, a task-aligned focal loss (TFL) which combines the classification difficulty with the regression loss is proposed, and TFL assigns lager weights to the hard instances and optimizes the classification and regression branches simultaneously. The ablation study proves the effectiveness of scene classification branch, TFL and their combination. The comparisons with 15 and 14 OOD methods on the DOTA and DIOR-R datasets validate the superiority of our method. Full article

Beach ridges originate from various depositional processes and occur in a variety of settings. This paper assesses their nature and distribution on the islands of the Greater Caribbean based on a literature review and the identification of sites using Google Earth© 7.3 imagery. The morphological and orientation parameters were measured for each site, and a measure of storm density was developed. These were statistically analysed to develop a classification of beach ridge types. The results show a diversity of beach ridge systems, in terms of setting, morphology, composition and preservation. The presence or absence of an adjacent coral reef is a major differentiating element at the regional level. A regional beach ridge plain classification is proposed, including two main classes, marine beach ridges and river-associated beach ridges, with further sub-divisions based on exposure to hurricanes or hurricanes plus swell waves. Full article

Accurate human posture classification in images and videos is crucial for automated applications across various fields, including work safety, physical rehabilitation, sports training, or daily assisted living. Recently, multimodal learning methods, such as Contrastive Language-Image Pretraining (CLIP), have advanced significantly in jointly understanding images and text. This study aims to assess the effectiveness of CLIP in classifying human postures, focusing on its application in yoga. Despite the initial limitations of the zero-shot approach, applying transfer learning on 15,301 images (real and synthetic) with 82 classes has shown promising results. The article describes the full procedure for fine-tuning, including the choice for image description syntax, models and hyperparameters adjustment. The fine-tuned CLIP model, tested on 3826 images, achieves an accuracy of over 85%, surpassing the current state-of-the-art of previous works on the same dataset by approximately 6%, its training time being 3.5 times lower than what is needed to fine-tune a YOLOv8-based model. For more application-oriented scenarios, with smaller datasets of six postures each, containing 1301 and 401 training images, the fine-tuned models attain an accuracy of 98.8% and 99.1%, respectively. Furthermore, our experiments indicate that training with as few as 20 images per pose can yield around 90% accuracy in a six-class dataset. This study demonstrates that this multimodal technique can be effectively used for yoga pose classification, and possibly for human posture classification, in general. Additionally, CLIP inference time (around 7 ms) supports that the model can be integrated into automated systems for posture evaluation, e.g., for developing a real-time personal yoga assistant for performance assessment. Full article

Three-dimensional object detection involves estimating the dimensions, orientations, and locations of 3D bounding boxes. Intersection of Union (IoU) loss measures the overlap between predicted 3D box and ground truth 3D bounding boxes. The localization task uses smooth-L1 loss with IoU to estimate the object’s location, and the classification task identifies the object/class category inside each 3D bounding box. Localization suffers a performance gap in cases where the predicted and ground truth boxes overlap significantly less or do not overlap, indicating the boxes are far away, and in scenarios where the boxes are inclusive. Existing axis-aligned IoU losses suffer performance drop in cases of rotated 3D bounding boxes. This research addresses the shortcomings in bounding box regression problems of 3D object detection by introducing an Improved Intersection Over Union (IIoU) loss. The proposed loss function’s performance is experimented on LiDAR-based and Camera-LiDAR-based fusion methods using the KITTI dataset. Full article

Human action recognition is a computer vision task that identifies how a person or a group acts on a video sequence. Various methods that rely on deep-learning techniques, such as two- or three-dimensional convolutional neural networks (2D-CNNs, 3D-CNNs), recurrent neural networks (RNNs), and vision transformers (ViT), have been proposed to address this problem over the years. Motivated by the fact that most of the used CNNs in human action recognition present high complexity, and the necessity of implementations on mobile platforms that are characterized by restricted computational resources, in this article, we conduct an extensive evaluation protocol over the performance metrics of five lightweight architectures. In particular, we examine how these mobile-oriented CNNs (viz., ShuffleNet-v2, EfficientNet-b0, MobileNet-v3, and GhostNet) execute in spatial analysis compared to a recent tiny ViT, namely EVA-02-Ti, and a higher computational model, ResNet-50. Our models, previously trained on ImageNet and BU101, are measured for their classification accuracy on HMDB51, UCF101, and six classes of the NTU dataset. The average and max scores, as well as the voting approaches, are generated through three and fifteen RGB frames of each video, while two different rates for the dropout layers were assessed during the training. Last, a temporal analysis via multiple types of RNNs that employ features extracted by the trained networks is examined. Our results reveal that EfficientNet-b0 and EVA-02-Ti surpass the other mobile-CNNs, achieving comparable or superior performance to ResNet-50. Full article

To minimize dependency on the availability of data labels, some WiFi-CSI based-gesture recognition solutions utilize an unsupervised representation learning phase prior to fine-tuning downstream task classifiers. In this case, however, the overall performance of the solution is negatively affected by domain factors present in the WiFi-CSI data used by the pre-training models. To reduce this negative effect, we propose an integration of the adversarial domain classifier in the pre-training phase. We consider this as an effective step towards automatic domain discovery during pre-training. We also experiment with multi-class and label versions of domain classification to improve situations, in which integrating a multi-class and single label-based domain classifier during pre-training fails to reduce the negative impact domain factors have on overall solution performance. For our extensive random and leave-out domain factor cross-validation experiments, we utilise (i) an end-to-end and unsupervised representation learning baseline, (ii) integration of both single- and multi-label domain classification, and (iii) so-called domain-aware versions of the aformentioned unsupervised representation learning baseline in (i) with two different datasets, i.e., Widar3 and SignFi. We also consider an input sample type that generalizes, in terms of overall solution performance, to both aforementioned datasets. Experiment results with the Widar3 dataset indicate that multi-label domain classification reduces domain shift in position (1.2% mean metric improvement and 0.5% variance increase) and orientation (0.4% mean metric improvement and 1.0% variance decrease) in domain factor leave-out cross-validation experiments. The results also indicate that domain shift reduction, when considering single- or multi-label domain classification during pre-training, is negatively impacted when a large proportion of negative view combinations contain views that originate from different domains within a substantial amount of mini-batches considered during pre-training. This is caused by the view contrastive loss repelling the aforementioned negative view combinations, eventually causing more domain shift in the intermediate feature space of the overall solution. Full article