GEABTM: A Generative Adversarial Network Ensemble Activation-Enabled Bi-directional Long Short-Term Memory Model for Automated Fruit Quality Detection

This study develops a data acquisition system for monitoring, detecting, and forecasting electrical energy consumption to support efficient energy management. Electrical parameters such as voltage, current, and power are measured using a PM2120 power meter via Modbus RTU RS485 and processed by an ESP32 microcontroller. The data are displayed in real-time through a Nextion Human-Machine Interface (HMI) and utilized as input for a Long Short-Term Memory (LSTM) model trained on historical consumption data. Safety features include LED indicators that activate when current reaches 80% of maximum capacity and a buzzer that signals threshold violations. Experimental results demonstrate high prediction accuracy, with RMSE values of 0.38 kW (5.32%) for phase R, 0.47 kW (7.55%) for phase S, and 0.28 kW (5.39%) for phase T. Transmission latency averages two to three seconds, while prediction computation is under 10 seconds. The system effectively reflects consumption trends, making it a reliable decision-support tool for enhancing energy efficiency in small- to medium-scale installations.

Continuous kinematics estimation methods are important in human-machine interaction systems because they offer a more natural and intuitive result than pattern-recognition methods. The estimation of upper-limb movement is important, which involves the estimation of the movement of the elbow and shoulder joints. However, the deformations of muscles around the elbow and shoulder can cause the shifting of the angle sensors. Nevertheless, Vicon can measure joint angles using cameras that have no contact with muscles. In this paper, we propose a Linear-Attention-based model (LABD) rather than other deep learning models to estimate upper-limb movements. The sEMG signals were collected from eight sEMG sensors and angles were measured from Vicon. The experimental results of LABD were compared with multilayer perceptrons (MLP), Temporal Convolutional Network (TCN), long short-term memory network (LSTM), and a dot-product attention-based model (DABD). The Pearson correlation coefficients (PCC) between the target and estimated joint angle sequences were calculated to evaluate the performance of each model. The Wilcoxon signed-rank results showed that the LABD significantly outperformed the other models.

Travel planning is a process that allows users to obtain maximum benefit from their time, cost and energy. When planning a route from one place to another, it is an important option to present alternative travel areas on the route. This study proposes a travel route planning (TRP) architecture using a Long Short-Term Memory (LSTM) and Extreme Gradient Boosting (XGBoost) model to improve both travel efficiency and environmental sustainability in route selection. This model incorporates carbon emissions directly into the route planning process by unifying user preferences, location recommendations, route optimization, and multimodal vehicle selection within a comprehensive framework. By merging environmental sustainability with user-focused travel planning, it generates personalized, practical, and low-carbon travel routes. The carbon emissions observed with TRP’s artificial intelligence (AI) recommendation route are presented comparatively with those of the user-determined route. XGBoost, Random Forest (RF), Categorical Boosting (CatBoost), Light Gradient Boosting Machine (LightGBM), (Extra Trees Regressor) ETR, and Multi-Layer Perception (MLP) models are applied to the TRP model. LSTM is compared with Recurrent Neural Networks (RNNs) and Gated Recurrent Unit (GRU) models. Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Squared Error (MSE), and Normalized Root Mean Square Error (NRMSE) error measurements of these models are carried out, and the best result is obtained using XGBoost and LSTM. TRP enhances environmental responsibility awareness within travel planning by integrating sustainability-oriented parameters into the decision-making process. Unlike conventional reservation systems, this model encourages individuals and organizations to prioritize eco-friendly options by considering not only financial factors but also environmental and socio-cultural impacts. By promoting responsible travel behaviors and supporting the adoption of sustainable tourism practices, the proposed approach contributes significantly to the broader dissemination of environmentally conscious travel choices.

With thousands of new websites emerging daily, distinguishing between legitimate and malicious web pages has become increasingly challenging, as many of these sites compromise users' private data without consent, posing severe cybersecurity threats. The absence of robust detection mechanisms exposes users to cyberattacks, financial fraud, and identity theft. While several Machine Learning (ML)-based techniques exist, they suffer from limitations such as reliance on handcrafted features and difficulty in adapting to evolving attack patterns. To mitigate these challenges, this paper introduces a fully automated deep learning (DL) based framework designed for the detection of malicious Uniform Resource Locators (URLs). The framework utilizes Large Language Models (LLMs) to generate high-quality URL embeddings that capture complex patterns and token relationships in URLs without manual feature engineering. These embeddings are then classified into four categories, i.e., defacement, malware, benign, and phishing, using a customized DL-based model that is finalized using extensive ablation experiments. The proposed DL model uses Long Short Term Memory (LSTM) and Gated Recurrent Unit (GRU) layers to capture long-range dependencies between the embeddings. The proposed system achieved the highest accuracy of 97.5% using a Bidirectional Encoder Representations from Transformers (BERT) and a DL-based model. With only 0.5M parameters, the BERT + DL model can classify samples in 0.119 ms. Additionally, to enhance interpretability and trustworthiness, the eXplainable AI (XAI) technique called Local Interpretable Model-Agnostic Explanations (LIME) is used to visualize model decisions to ensure the model's transparency and reliability in a real-time setting.

Objective.Accurate dose calculation on cone beam computed tomography (CBCT) images is essential for modern proton treatment planning workflows, particularly when accounting for inter-fractional anatomical changes in adaptive treatment scenarios. Traditional CBCT-based dose calculation suffers from image quality limitations, requiring complex correction workflows. This study develops and validates a deep learning approach for direct proton dose calculation from CBCT images using extended long short-term memory (xLSTM) neural networks.Approach.A retrospective dataset of 40 head-and-neck cancer patients with paired planning CT and treatment CBCT images was used to train an xLSTM-based neural network (CBCT-NN). The architecture incorporates energy token encoding and beam's-eye-view sequence modeling to capture spatial dependencies in proton dose deposition patterns. Training utilized 82 500 paired proton pencil beam configurations with Monte Carlo (MC)-generated ground truth doses. Validation was performed on five independent patients using gamma analysis, mean percentage dose error (MPDE) assessment, and dose-volume histogram comparison.Main results.The CBCT-NN achieved gamma pass rates of 95.1 ± 2.7% using 2 mm/2% criteria. MPDEs were 2.6 ± 1.4% in high-dose regions (>90% of max dose) and 5.9 ± 1.9% globally. Dose-volume histogram analysis showed excellent preservation of target coverage metrics (clinical target volume V95% difference: -0.6 ± 1.1%) and organ-at-risk constraints (parotid mean dose difference: -0.5 ± 1.5%). Computation time is under 3 min without sacrificing MC-level accuracy.Significance.This study demonstrates the proof-of-principle of direct CBCT-based proton dose calculation using xLSTM neural networks. The approach eliminates traditional correction workflows while achieving comparable accuracy and computational efficiency suitable for adaptive protocols.

This paper addresses the challenges of dynamic environments and multimodal data fusion in multimodal transport path optimization for smart ports by proposing a GL-SSL Model that integrates Graph Neural Networks (GCN), Long Short-Term Memory (LSTM), and Self-Supervised Learning (SSL). The model fully exploits the graph-structured information of port transport networks and their temporal variations, while SSL enhances feature representation, enabling efficient optimization of path planning. Experiments were conducted on multiple public datasets, including AIS data from the Port of Rotterdam, global shipping data, and port net revenue data. Results show that the GL-SSL Model achieved significant improvements in key performance metrics. Specifically, the optimized path length reached 80 km, the transport cost was reduced to 200 cost-units (a composite metric reflecting fuel consumption, equipment wear, and labor cost), and the delay rate was maintained at 0.05 (5%), all of which are substantially better than traditional algorithms and other deep learning models. Furthermore, the model demonstrated stable performance under complex scenarios such as peak traffic, adverse weather, and equipment failures, with rapid convergence of training loss and strong robustness. These findings highlight the model’s adaptability and practical application potential. Overall, this work provides effective technical support for multimodal transport path optimization in smart ports and carries important theoretical significance and broad application prospects.

The Corsi Block Tapping Test (CBT) assesses short-term visuospatial memory, while the Walking Corsi Test (WalCT) introduces greater motor and spatial demands. The impact of visual trajectory presentation on these tests remains debated. To examine the effect of visual presentation of the trajectory on visuospatial short-term memory and topographic memory during the CBT and WalCT. A total of 37 students completed the Corsi task paradigm under two conditions: CBT and WalCT. Both were performed in classical versions (with visual trajectory presentation) and automated versions (without visual trajectory presentation). Each test was conducted in forward and backward modalities. Span and Total Product values were recorded. Sex, age and physical activity level were considered in the analysis. For Span, only the type of test influenced performance, with higher results observed in the CBT (7.29 ± 1.13) compared to the WalCT (6.18 ± 1.55) (p < 0.001). Regarding Total Product, significant effects were found for both the type of test (CBT = 84.83 ± 26.21; WalCT = 58.76 ± 28.99; p = 0.026) and the modality (Forward = 74.85 ± 31.56; Backward = 68.74 ± 29.23; p = 0.026). Age significantly interacted as a covariate in both analyses (p < 0.001). The visual presentation of the trajectory did not improve performance. However, the type and modality of the test directly influenced final performance. Additionally, age emerged as a factor affecting performance in the Corsi paradigms, while physical activity level and sex showed no significant effects.

Schizophrenia is a complex and chronic psychiatric disorder that affects millions worldwide, significantly impairing cognitive and emotional functioning. Early detection is crucial for improving patient outcomes, yet traditional diagnostic approaches remain subjective and often delayed. Recent advances in Artificial Intelligence (AI) and Machine Learning (ML) have paved the way for automated and more accurate schizophrenia detection. This review explores various ML techniques, including traditional algorithms such as Support Vector Machines (SVM) and Random Forests, as well as deep learning models like Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks. The study delves into different data modalities, including neuroimaging (MRI, fMRI), Electroencephalography (EEG), and clinical assessments, highlighting their role in early schizophrenia identification. Performance evaluation metrics such as accuracy, sensitivity, specificity, and the Area Under the Curve (AUC) are analyzed across multiple studies. Challenges such as data heterogeneity, small sample sizes, and ethical concerns are discussed alongside potential solutions, including explainable AI and multimodal integration. Future directions emphasize real-time monitoring, AI-powered mobile applications, and the development of robust, generalizable models. This review underscores the transformative potential of ML in mental health diagnostics and its role in revolutionizing schizophrenia detection and intervention strategies. Major Findings: This review highlight that Machine Learning (ML) algorithms, particularly support vector machines, Random Forests, CNNs, and LSTMs, have demonstrated promising accuracy in early schizophrenia detection. Neuroimaging, EEG, and clinical assessments serve as key data sources, with multimodal approaches enhancing diagnostic precision. Performance metrics indicate strong potential, though variability exists due to data quality and sample size. Challenges such as data heterogeneity and ethical concerns persist, but advances in explainable AI and integration of diverse data types offer solutions. The future lies in real-time, mobile, and generalizable AI models that support early, accessible, and objective mental health interventions.

A hybrid power load forecasting model based on evolutionary strategy and long short term memory

Meteorological drought severity forecasting utilizing blended modelling.

Tracking is now popular in real world. Precise tracking of objects in real-time videos is a challenging task. With billions of fans, football is a rapidly expanding sport that has proven essential to many nations and their citizens in particular. None of the numerous great target tracking algorithms have surfaced in recent years primarily deep learning and correlation filtering that can track players in soccer game videos with high accuracy. In this paper, the proposed system is used You Only Look Once version 8-nano (YOLOv8n) for Multi-Object Detection (MOD) to get higher detection accuracy results. Moreover, this system is based on the hybrid method for tracking. The hybrid method is combined with stacked Long Short Term Memory (LSTM) and Fairness of Detection and Re-identification in Multipe Object Tracking (FairMOT). The experimental analysis shows that the proposed system is efficiently and better accuacy because the best detection results with YOLOv8n is 93% for precision, 91% for recall and 92% for mAP(50) with own dataset. After using the proposed system, the average of the Multi Object Tracking Accuracy (MOTA) is 80 % at IoU-Threshold 0.5, the average of the Multi Object Tracking Precision (MOTP) is 89% at IoU-Threshold 0.8 and the average of the final mAP is 96% at IoU- Threshold 0.5 by using hybrid method for tracking.

Cardiotocography (CTG) is a widely used technique for fetal monitoring. This study presents CTGFusionNet, a novel multimodal adaptive framework designed for prenatal analysis. The framework integrates attention-based adaptive Bi-Directional Convolutional Neural Networks (Bi-CNN) with Long Short-Term Memory (LSTM) networks to improve the accuracy of fetal distress prediction. The methodology begins with an initial data preprocessing phase, followed by signal segmentation and enhancement. Thereafter, the FHR and UC signals are transformed into two-dimensional representations using embedding layers and subsequently integrated through concatenation. The spatial features of the synchronized signals are extracted using the proposed adaptive Bi-CNN. Multi-head attention is then applied to emphasize the most relevant information, and the temporal features are captured using an LSTM network. In the final stage, the most relevant features from the perinatal clinical data are identified using the Relief, Lasso, and Information Gain algorithms and then integrated with the processed signals. Furthermore, classification results are obtained using a fully connected layer and sigmoid function. The results demonstrate that CTGFusionNet leads to significant improvements in performance measures, namely accuracy, sensitivity, and specificity, with values of 97.85%, 97.07%, and 98.65%, respectively. This suggests that CTGFusionNet-a multimodal approach that combines FHR, UC, and clinical data, provides a more reliable and precise method for the early detection and prediction of fetal distress. The proposed approach has the potential to significantly improve prenatal care outcomes by enabling accurate interventions.

Emotion recognition (ER) is crucial for understanding human behaviours, social interactions, and psychological well-being. Electroencephalography (EEG) has emerged as a promising tool for capturing the neural correlates of emotions. This work is a systematic review of articles in ER using EEG signals. A total of 120 articles from 1041 articles were selected based on PRISMA guidelines using defined inclusion and exclusion criteria, published between 2018 and 2024. This article aims to provide an in-depth understanding of the current landscape of ER from EEG signals utilizing deep learning (DL). This review offers valuable guidance for researchers and practitioners seeking more refined and reliable emotion classification systems. To explore the effectiveness of DL models in EEG-based ER, several potential DL models, such as convolutional neural network, long short-term memory (LSTM), gated recurrent unit (GRU), hybrid bidirectional LSTM (BiLSTM), bidirectional GRU, and advanced DL models such as convolutional recurrent neural network and EEG-Conformer models are applied to two popular datasets, SEED and GAMEEMO, respectively, to depict the full process of ER. Additionally, the performance of DL models is also compared with the performance of basic machine learning (ML) models such as SVM, k-nearest neighbors, logistic regression, and boosting algorithms such as AdaBoost, XGBoost and LightGBM. Through extensive experiments and performance evaluations, the performance of different models when applied to the datasets mentioned above is compared. The accuracy, precision, recall, and F1-scores are analysed to determine the most effective model for EEG-based ER. The findings of this study demonstrate that the performance of hybrid DL models is more efficacious than that of ML models. The best-performing model (BiLSTM) classified the emotions, with an accuracy of 90.54% when applied to the GAMEEMO dataset. This research contributes to the growing body of literature on ER and provides insights into the feasibility of using EEG signals to understand emotional states, and presents a structured roadmap for future exploration. The findings can aid in the development of more accurate and reliable ER systems, which can have wide-ranging applications in psychology, social sciences, and human-computer interactions.

Various industries, such as healthcare and surveillance, depend heavily on the ability to recognize human activity. The “human activity recognition (HAR) using smartphones data set” can be found in the UCI online repository and includes accelerometer and gyroscope readings recorded during a variety of human activities. The accelerometer and gyroscope signals are also subjected to a band-pass filter to eliminate unwanted frequencies and background noise. This method effectively decreases the dimensionality of the feature space while improving the model's accuracy and efficiency. “Convolutional neural networks (CNNs)” and “long shortterm memory (LSTM)” networks are combined to create pyramidal dilated convolutional memory network (PDCMN), which is the final proposal. Results from experiments demonstrate the effectiveness and reliability of the suggested method, demonstrating its potential for precise and effective HAR in actuality schemes.

Are Vision-xLSTM-embedded U-Nets better at segmenting medical images?

The rise of social media has enabled public expression but also fueled the spread of hate speech, contributing to social tensions and potential violence. Natural language processing (NLP), particularly text classification, has become essential for detecting hate speech. This study develops a hate speech detection model on Twitter using FastText with bidirectional long short-term memory (Bi-LSTM) and explores multilingual bidirectional encoder representations from transformers (M-BERT) for handling diverse languages. Data augmentation techniques-including easy data augmentation (EDA) methods, back translation, and generative adversarial networks (GANs)-are employed to enhance classification, especially for imbalanced datasets. Results show that data augmentation significantly boosts performance. The highest F1-scores are achieved by random insertion for Indonesian (F1-score: 0.889, Accuracy: 0.879), synonym replacement for English (F1-score: 0.872, Accuracy: 0.831), and random deletion for German (F1-score: 0.853, Accuracy: 0.830) with the FastText + Bi-LSTM model. The M-BERT model performs best with random deletion for Indonesian (F1-score: 0.898, Accuracy: 0.880), random swap for English (F1 score: 0.870, Accuracy: 0.866), and random deletion for German (F1-score: 0.662, Accuracy: 0.858). These findings underscore that data augmentation effectiveness varies by language and model. This research supports efforts to mitigate hate speech’s impact on social media by advancing multilingual detection capabilities.

Beyond predictive accuracy: Statistical validation of feature importance in biomedical machine learning.

Fusion of transfer learning models for detection of alzheimer's disease using bidirectional long short-term memory with equilibrium optimization algorithm.

In the modern world, people worldwide face different forms of depression due to factors such as workplace stress, economic pressures, and other causes. The rise of Artificial Intelligence (AI) has enabled data analysis and solving of real-world problems. People frequently use social media platforms to communicate and express their feelings. Hence, social media data is helpful for research purposes, particularly for automatic depression detection. Numerous scholarly works have explored using learning-based approaches to identify sadness from social media interactions. However, individual existing deep learning models have limitations, such as the inability to capture contextual and sequential dependencies in text fully. We addressed this by proposing a deep learning-based, non-invasive approach to identify depression in social media conversations. Our proposed approach involves a novel hybrid deep learning model, Depression Detection Network (DDNet), which combines Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) models. The model was trained and tested on a manually annotated dataset of 8500 depression-related tweets (6,800 for training and 1,700 for testing) collected via the Twitter Application Programming Interface (API). The DDNet model achieved a high accuracy of 96.21%, outperforming baseline models such as standalone LSTM (92.31%) and Recurrent Neural Network (RNN) (91.43%). Furthermore, we developed Hybrid Deep Learning-based Depression Detection (HDL-DD), an algorithm that processes social media text and predicts potential depressive tendencies. The experimental results indicate that DDNet significantly improves depression classification, achieving 95% precision, 96% recall, and 95% F1-score, demonstrating its effectiveness over existing methods. By recognizing depression with a 96.21% accuracy rate, our deep learning model outperformed previous state-of- the-art approaches, making it a promising tool for automated depression monitoring applications. This approach could be integrated into real-world social media-based mental health monitoring applications, supporting early intervention efforts and contributing to AI-driven healthcare solutions.