High Accuracy Training Research Articles

Enabling Intelligence on the Edge: Leveraging Edge Impulse to Deploy Multiple Deep Learning Models on Edge Devices for Tomato Leaf Disease Detection

Tomato diseases, including Leaf blight, Leaf curl, Septoria leaf spot, and Verticillium wilt, are responsible for up to 50% of annual yield loss, significantly impacting global tomato production, valued at approximately USD 87 billion. In Ghana, there is a yield gap of about 50% in tomato production, which requires drastic measures to increase the yield of tomatoes. Conventional diagnostic methods are labor-intensive and impractical for real-time application, highlighting the need for innovative solutions. This study addresses these issues in Ghana by utilizing Edge Impulse to deploy multiple deep-learning models on a single mobile device, facilitating the rapid and precise detection of tomato leaf diseases in the field. This work compiled and rigorously prepared a comprehensive Ghanaian dataset of tomato leaf images, applying advanced preprocessing and augmentation techniques to enhance robustness. Using TensorFlow, we designed and optimized efficient convolutional neural network (CNN) architectures, including MobileNet, Inception, ShuffleNet, Squeezenet, EfficientNet, and a custom Deep Neural Network (DNN). The models were converted to TensorFlow Lite format and quantized to int8, substantially reducing the model size and improving inference speed. Deployment files were generated, and the Edge Impulse platform was configured to enable multiple model deployments on a mobile device. Performance evaluations across edge hardware provided metrics such as inference speed, accuracy, and resource utilization, demonstrating reliable real-time detection. EfficientNet achieved a high training accuracy of 97.12% with a compact 4.60 MB model size, proving its efficacy for mobile device deployment. In contrast, the custom DNN model is optimized for microcontroller unit (MCU) deployment. This edge artificial intelligence (AI) technology integration into agricultural practices offers scalable, cost-effective, and accessible solutions for disease classification, enhancing crop management, and supporting sustainable farming practices.

Remote sensing image road network detection based on channel attention mechanism

Extracting and detecting road network consistency from high-resolution remote sensing images has been a hot and difficult problem in the computer vision. Although it has made significant progress, there is still a phenomenon of high training accuracy but unsatisfactory actual extraction and detection results. The attention mechanism is one of the efficient and practical mechanisms in deep learning. It improves the performance of deep learning by selectively focusing on a portion of all information while ignoring other visible information, while effectively utilizing computing resources. Numerous experiments have also confirmed that the attention mechanism is resource-saving and effective. Its plug and play feature brings great convenience to programmers. In order to provide better road network detection results and solve the above problem, this paper combines the channel attention mechanism with ResNet and proposes SE-ResNet and ECA-ResNet for remote sensing image road network detection, making networks extract and learn road network features and ignore some non-road network features. The experimental results show that on the Massachusetts roads (MR) and CHN6-CUG roads datasets, ECA-ResNet and SE-ResNet based on channel attention mechanism perform similar to LeNet7 and ResNet in terms of accuracy, loss, accuracy convergence, and loss convergence, and even increase a certain computational burden. However, their final road network detection results (including road network detection pixel count, precision, recall, accuracy, IOU, F1 score, and actual road network detection result) of the former are significantly better than those of the latter. The channel attention mechanism makes the deep neural network pay more attention to the extraction and learning of road network features, while ignoring the extraction and learning of some non-road network features, which improves the accuracy of containing road network samples and reduces the accuracy of not containing road network samples. Therefore, the performance of ECA-ResNet and SE-ResNet is similar to that of LeNet7 and ResNet in the accuracy, loss, accuracy convergence and loss convergence, but the final road network detection results of ECA-ResNet and SE-ResNet are significantly better than those of LeNet7 and ResNet. Therefore, the proposed ECA-ResNet and SE-ResNet have broad application prospects in road network detection, especially ECA-ResNet.

Groundwater LNAPL Contamination Source Identification Based on Stacking Ensemble Surrogate Model

Groundwater LNAPL (Light Non-Aqueous Phase Liquid) contamination source identification (GLCSI) is essential for effective remediation and risk assessment. Addressing the GLCSI problem often involves numerous repetitive forward simulations, which are computationally expensive and time-consuming. Establishing a surrogate model for the simulation model is an effective way to overcome this challenge. However, how to obtain high-quality samples for training the surrogate model and which method should be used to develop the surrogate model with higher accuracy remain important questions to explore. To this end, this paper innovatively adopted the quasi-Monte Carlo (QMC) method to sample from the prior space of unknown variables. Then, this paper established a variety of individual machine learning surrogate models, respectively, and screened three with higher training accuracy among them as the base-learning models (BLMs). The Stacking ensemble framework was utilized to integrate the three BLMs to establish the ensemble surrogate model for the groundwater LNAPL multiphase flow numerical simulation model. Finally, a hypothetical case of groundwater LNAPL contamination was designed. After evaluating the accuracy of the Stacking ensemble surrogate model, the differential evolution Markov chain (DE-MC) algorithm was applied to jointly identify information on groundwater LNAPL contamination source and key hydrogeological parameters. The results of this study demonstrated the following: (1) Employing the QMC method to sample from the prior space resulted in more uniformly distributed and representative samples, which improved the quality of the training data. (2) The developed Stacking ensemble surrogate model had a higher accuracy than any individual surrogate model, with an average R2 of 0.995, and reduced the computational burden by 99.56% compared to the inversion process based on the simulation model. (3) The application of the DE-MC algorithm effectively solved the GLCSI problem, and the mean relative error of the identification results of unknown variables was less than 5%.

Modeling of hydrogen separation through Pd membrane with vacuum pressure using Taguchi and machine learning methods

The performance of hydrogen purification using palladium (Pd) membrane is analyzed by Taguchi and machine learning (ML) methods. Three system factors are considered: the feed gas mixture composition (i.e., H2, CO2, and H2O), retentate-side total pressure, and vacuum pressure. The Taguchi method designs the experimental cases based on the constructed orthogonal array. In addition, the hydrogen flux is investigated by the analyses of variance (ANOVA) and artificial neural networks (ANN) methods. The results show that the effects of the considered factors on the hydrogen permeation performance can be ranked as feed gas mixture composition > retentate-side total pressure > vacuum pressure. Both the retentate-side pressure and vacuum pressure present a positive effect on hydrogen flux. The average relative error of hydrogen flux between the experimental results and predictions by ANN is 2.1%, which proves that the ANN method is an effective technique for predicting the hydrogen flux through the Pd membrane. The ML classification test is then performed for hydrogen purity by three machine learning methods: decision tree (DT), support vector machine (SVM), and ensemble method. Among the various classification models, the Quadratic SVM model exhibits a relatively high average training accuracy but shows overfitting. However, the bagged model of the ensemble method can achieve an impressive training accuracy of 91.4% and a prediction accuracy of 85.7%.

Deep Learning based Apple Fruit Disease Detection using Dense Net Recursive Convolutional Neural Network (DNRCNN)

Every year, fruit diseases cost the apple industry a lot of money. It can be challenging for growers to identify various apple infections because the symptoms of various illnesses are often similar and may overlap. In this study, we suggest a deep learning-based method for identifying and categorizing apple diseases. Dataset generation, which includes data collection and data labelling, is the first stage of the investigation. On the prepared dataset, we then trained a deep learning-based Dense Net Recurrent Convolutional Neural Network (DNRCNN) model for automatically classifying apple diseases. The end-to-end learning algorithm DNRCNN is appropriate for a range of tasks including image classification, object detection, and segmentation because it automatically extracts complex features from source images and learns them directly. Initialize the parameters of the proposed deep model using transfer learning. To avoid over-fitting, data augmentation techniques like rotation, translation, reflection, and scaling are also used. On the prepared dataset, the proposed Dense Net Recursive Convolutional Neural Network (DNRCNN) model achieves promising results, with an accuracy of about 96%. Some of the intricate and helpful image characteristics for detection are captured by the suggested model for classification. The model can learn the higher-order features of two adjacent layers that are not in the same channel but have a high correlation more effectively than existing techniques. High training and validation accuracy have been achieved when training and validating the suggested model. The findings support the method's usefulness in categorizing different apple diseases and show that it can be useful for farmers

Evolutionary Strategies Enable Systematic and Reliable Uncertainty Quantification: A Proof-of-Concept Pilot Study on Resting-State Functional MRI Language Lateralization.

Reliable and trustworthy artificial intelligence (AI), particularly in high-stake medical diagnoses, necessitates effective uncertainty quantification (UQ). Existing UQ methods using model ensembles often introduce invalid variability or computational complexity, rendering them impractical and ineffective in clinical workflow. We propose a UQ approach based on deep neuroevolution (DNE), a data-efficient optimization strategy. Our goal is to replicate trends observed in expert-based UQ. We focused on language lateralization maps from resting-state functional MRI (rs-fMRI). Fifty rs-fMRI maps were divided into training/testing (30:20) sets, representing two labels: "left-dominant" and "co-dominant." DNE facilitated acquiring an ensemble of 100 models with high training and testing set accuracy. Model uncertainty was derived from distribution entropies over the 100 model predictions. Expert reviewers provided user-based uncertainties for comparison. Model (epistemic) and user-based (aleatoric) uncertainties were consistent in the independently and identically distributed (IID) testing set, mainly indicating low uncertainty. In a mostly out-of-distribution (OOD) holdout set, both model and user-based entropies correlated but displayed a bimodal distribution, with one peak representing low and another high uncertainty. We also found a statistically significant positive correlation between epistemic and aleatoric uncertainties. DNE-based UQ effectively mirrored user-based uncertainties, particularly highlighting increased uncertainty in OOD images. We conclude thatDNE-based UQ correlates with expert assessments, making it reliable for our use case and potentially for other radiology applications.

Convolutional Neural Network‐Based CT Image Segmentation of Kidney Tumours

ABSTRACTKidney tumours are one of the most common tumours in humans and the main current treatment is surgical removal. The CT images are usually manually segmented by a specialist for pre‐operative planning, but this can be influenced by the surgeon's experience and skill and can be time‐consuming. Due to the complex lesions and different morphologies of kidney tumours that make segmentation difficult, this article proposes a convolutional neural network‐based automatic segmentation method for CT images of kidney tumours to address the most common problems of boundary blurring and false positives in tumour segmentation images. The method is highly accurate and reliable, and is used to assist doctors in surgical planning as well as diagnostic treatment, relieving medical pressure to a certain extent. The EfficientNetV2‐UNet segmentation model proposed in this article includes three main parts: feature extractor, reconstruction network and Bayesian decision algorithm. Firstly, for the phenomenon of tumour false positives, the EfficientNetV2 feature extractor, which has high training accuracy and efficiency, is selected as the backbone network, which extracts shallow features such as tumour location, morphology and texture in the CT image by downsampling. Secondly, on the basis of the backbone network, the reconstruction network is designed, which mainly consists of conversion block, deconvolution block, convolution block and output block. Then, the up‐sampling architecture is constructed to gradually recover the spatial resolution of the feature map, fully identify the contextual information and form a complete encoding–decoding structure. Multi‐scale feature fusion is achieved by superimposing all levels of feature map channels on the left and right sides of the network, preventing the loss of details and performing accurate tumour segmentation. Finally, a Bayesian decision algorithm is designed for the edge blurring phenomenon of segmented tumours and cascaded over the output of the reconstruction network, combining the edge features of the original CT image and the segmented image for probability estimation, which is used to improve the accuracy of the model edge segmentation. Medical images in NII special format were converted to Numpy matrix format using python, and then more than 2000 CT images containing only kidney tumours were selected from the KiTS19 dataset as the dataset for the model, and the dimensions were standardised to 128 × 128, and the experimental results show that the model outperforms many other advanced models with good segmentation performance.

Predicting Therapeutic Response to Hypoglossal Nerve Stimulation Using Deep Learning.

To develop and validate machine learning (ML) and deep learning (DL) models using drug-induced sleep endoscopy (DISE) images to predict the therapeutic efficacy of hypoglossal nerve stimulator (HGNS) implantation. Patients who underwent DISE and subsequent HGNS implantation at a tertiary care referral center were included. Six DL models and five ML algorithms were trained on images from the base of tongue (BOT) and velopharynx (VP) from patients classified as responders or non-responders as defined by Sher's criteria (50% reduction in apnea-hypopnea index (AHI) and AHI < 15 events/h). Precision, recall, F1 score, and overall accuracy were evaluated as measures of performance. In total, 25,040 images from 127 patients were included, of which 16,515 (69.3%) were from responders and 8,262 (30.7%) from non-responders. Models trained on the VP dataset had greater overall accuracy when compared to BOT alone and combined VP and BOT image sets, suggesting that VP images contain discriminative features for identifying therapeutic efficacy. The VCG-16 DL model had the best overall performance on the VP image set with high training accuracy (0.833), F1 score (0.78), and recall (0.883). Among ML models, the logistic regression model had the greatest accuracy (0.685) and F1 score (0.813). Deep neural networks have potential to predict HGNS therapeutic efficacy using images from DISE, facilitating better patient selection for implantation. Development of multi-institutional data and image sets will allow for development of generalizable predictive models. N/A Laryngoscope, 2024.

Evaluating and Comparing AI Models for Hourly Energy Demand Prediction

This study investigates the application of various AI models to predict energy demand, comparing the performance of four specific models: Decision Tree, Random Forest, Gradient Boosting, and Linear Regression. The evaluation of these models' prediction performance reveals that ensemble methods like Random Forest and Gradient Boosting exhibit promising generalization capabilities, while the Decision Tree model shows high training accuracy but suffers from overfitting. The discussion underscores the importance of ensemble techniques and feature engineering optimization in mitigating overfitting and enhancing forecast accuracy. Furthermore, the study highlights the potential of AI-driven approaches to promote sustainability and resilience in energy systems, emphasizing the need for further optimization and collaboration among stakeholders to achieve a cleaner, more sustainable energy future.

Novel embedding model predicting the credit card's default using neural network optimized by harmony search algorithm and vortex search algorithm

In today's banking and financial system, using a credit card has become indispensable. The credit card industry has existed due to a shift in consumer preferences and a rise in national economic growth. The number of issuing banks, card issuers, and transaction volumes has increased significantly. Nevertheless, owing to the growth in the number of transactions made with credit cards, both the total amount due and the rate of defaults on credit card loans have become issues that cannot be neglected. This issue must be resolved to ensure the continued and prosperous growth of the banking industry in the years to come. Currently, a few optimization algorithms—Whale optimization algorithm (WOA), Harmony Search (HS), Multi-verse optimization (MVO), and Vortex Search (VS)—have been used to achieve this purpose. However, because credit card default data is volatile and unequal, it is challenging for typical optimization algorithms to offer steady approaches with optimal performance. Studies have indicated that optimizing algorithms with suitable properties can significantly improve performance. To improve performance, some tuning was applied to the ANN. This study will assess twenty-three parameters, and the efficacy of all four approaches will be compared using ROC and AUC evaluations. The suggested model's performance is contrasted with a scenario where the classifiers were trained using original data. In contrast, the AUC values for VS-MLP were 0.7407 and 0.7271, while those for HS-MLP were 0.7074 and 0.6997. In the training and testing phases, AUC values of 0.7469 and 0.7329 from MVO-MLP and 0.72 and 0.7185 from WOA-MLP, respectively. The results show that the training accuracy of HS, VSA, MVO, and WOA are similar; MVO has the highest training accuracy. The credit card industry can benefit significantly from this methodology, which may help resolve default probabilities.

Federated Learning with Efficient Aggregation via Markov Decision Process in Edge Networks

Federated Learning (FL), as an emerging paradigm in distributed machine learning, has received extensive research attention. However, few works consider the impact of device mobility on the learning efficiency of FL. In fact, it is detrimental to the training result if heterogeneous clients undergo migration or are in an offline state during the global aggregation process. To address this issue, the Optimal Global Aggregation strategy (OGAs) is proposed. The OGAs first models the interaction between clients and servers of the FL as a Markov Decision Process (MDP) model, which jointly considers device mobility and data heterogeneity to determine local participants that are conducive to global aggregation. To obtain the optimal client participation strategy, an improved σ-value iteration method is utilized to solve the MDP, ensuring that the number of participating clients is maintained within an optimal interval in each global round. Furthermore, the Principal Component Analysis (PCA) is used to reduce the dimensionality of the original features to deal with the complex state space in the MDP. The experimental results demonstrate that, compared with other existing aggregation strategies, the OGAs has the faster convergence speed and the higher training accuracy, which significantly improves the learning efficiency of the FL.

Comparative Study of Imbalanced Data Oversampling Techniques for Peer-to-Peer Landing Loan Prediction

Purpose: Data imbalances that often occur in the classification of loan data on the Peer-to-Peer Lending platform cancause algorithm performance to be less than optimal, causing the resulting accuracy to decrease. To overcome thisproblem, appropriate resampling techniques are needed so that the classification algorithm can work optimally andprovide results with optimal accuracy. This research aims to find the right resampling technique to overcome theproblem of data imbalance in data lending on peer-to-peer landing platforms.Methods: This study uses the XGBoost classification algorithm to evaluate and compare the resampling techniquesused. The resampling techniques that will be compared in this research include SMOTE, ADACYN, Border Line, andRandom Oversampling.Results: The highest training accuracy was achieved by the combination of the XGBoost model with the Boerder Lineresampling technique with a training accuracy of 0.99988 and the combination of the XGBoost model with the SMOTEresampling technique. In accuracy testing, the combination with the highest accuracy score was achieved by acombination of the XGBoost model with the SMOTE resampling technique.Novelty: It is hoped that from this research we can find the most suitable resampling technique combined with theXGBoost sorting algorithm to overcome the problem of unbalanced data in uploading data on peer-to-peer lendingplatforms so that the sorting algorithm can work optimally and produce optimal accuracy.

Wavelet entropy-based damage characterization and material phase differentiation in concrete using acoustic emissions

Damage evolution in concrete is a complex phenomenon involving micro-cracks that originate from different material phases (mortar matrix, ITZ and aggregate), and culminate into a major crack. The information about the individual damaged material phases can provide major insights into the global failure behaviour, cracking path, and fracture processes. The evolution of damage, damage mechanisms and the material phase of origin of micro-cracks is highly influenced by the type of concrete (Normal and high strength). In order to investigate the evolution of damage in concrete of varying strength, notched beam specimens with different water-to-cement ratios have been prepared and tested under CMOD control. The evolution and growth of damage have been continuously monitored during testing through acoustic emission techniques. Based on acoustic emission results, it has been demonstrated that the damage progression in concrete can be better described as a function of the wavelet entropy of the AE events occurring during the entire loading duration, and therefore, an analytical model for the damage evolution has been proposed. The study reveals that, under the high-strength category, the level of deterioration reduces as the w/c ratio rises, while in normal-strength concrete, a reverse trend is observed. Furthermore, a novel approach has been proposed that utilizes wavelet entropy and amplitude to differentiate amongst various damaged material phases with the help of an unsupervised clustering algorithm. Normal-strength concrete with a w/c ratio of 0.4 has been used to demonstrate the proposed methodology and has been validated with the help of experiments performed on the mortar and the parent stone of aggregate. A higher occurrence of mode-I cracks compared to mode-II cracks across all material phases has been observed for the specimen. Additionally, ITZ and aggregate exhibit a greater proportion of tensile and shear cracking events than the matrix phase. In the end, a deep learning convolutional neural network (CNN) model is devised utilizing labelled scalogram images of the AE waveforms to discriminate between the various damaged material phases. A 10-fold training cross-validation of the dataset has been carried out to achieve a stable performance. The deep learning model performs well in discriminating the damaged material phases with high training, validation, and testing accuracy.

GROUNDNUT CROP PEST DETECTION AND CLASSIFICATION USING COMPREHENSIVE DEEP-LEARNING MODELS

Pests pose a significant threat to crops, leading to substantial economic losses and decreased food production. Early detection and accurate classification of pests in crops are crucial for effective pest management strategies. In this study, we propose a method for pest detection and classification in groundnut crops using deep learning models. In this research, we compare the performance of three deep learning models, namely Custom CNN [proposed], LeNet-5, and VGG-16, for groundnut pest detection and classification. A comprehensive dataset containing images of diverse groundnut crop pests, including thrips, aphids, armyworms, and wireworms, from the IP102 dataset was utilized for model evaluation. The performance is evaluated using reliability metrics such as accuracy and loss. These findings demonstrate the utility of deep learning models for reliable pest classification of groundnut crops. The Custom CNN [proposed] model demonstrates high training accuracy but potential overfitting, while the VGG-16 model performs well on both training and test data, showcasing its ability to generalize. The models’ accuracy in predicting pest species underscores their capability to capture and utilize visual patterns for precise classification. These findings underscore the potential of deep learning models, particularly the VGG-16 model, for pest detection and classification in groundnut crops. The knowledge gained from this study can contribute to the development of practical pest management strategies and aid in maintaining crop health and productivity. Further analysis and comparisons with other models are recommended to comprehensively evaluate the competitiveness and suitability of deep learning models in real-world applications of pest detection and classification in agricultural settings.

Research on the method of diesel particulate filters carbon load recognition based on deep learning

Because the carbon load inside a diesel particulate filters (DPF) affects the DPF regeneration, and the carbon load recognition is significant for the particulate matter (PM) emission control. It is necessary to investigate an on-board DPF carbon load recognition method because the carbon load cannot be directly measured by sensors. Aiming to build a DPF carbon load prediction model adopting the deep learning method, this paper proposes a DPF carbon load identification model based on different experimental parameters using a layered one dimension convolutional neural network (1D-CNN) method. To improve data validity, this paper adopts two data-processing methods. The data pre-processing adopts data splicing method to complete the construction of the original sample set, and the data after-processing uses wavelet packet transform method to establish the feature sample sets. The model adopts the optimal feature dataset constructed by three input parameters, i.e., temperature difference, pressure difference, and exhaust mass flow, and has both high training accuracy and test accuracy above 90 %. The pressure difference is the most important influencing input parameter, and the three-parameter sample set (ΔT + ΔP + Q) has great recognition accuracy and good model stability with the high training accuracy and test accuracy as well as less iteration.

Feasibility Layer Aided Machine Learning Approach for Day-Ahead Operations

Day-ahead operation involves a complex and computationally intensive optimization process to determine the generator commitment schedule and dispatch. The optimization process is a mixed-integer linear program (MILP) also known as security-constrained unit commitment (SCUC). Independent system operators (ISOs) run SCUC daily and require state-of-the-art algorithms to speed up the process. Existing patterns in historical information can be leveraged for model reduction of SCUC, which can provide significant time savings. In this paper, machine learning (ML) based classification approaches, namely logistic regression, neural networks, random forest and K-nearest neighbor, were studied for model reduction of SCUC. The ML was then aided with a feasibility layer (FL) and post-process technique to ensure high quality solutions. The proposed approach is validated on several test systems namely, IEEE 24-Bus system, IEEE-73 Bus system, IEEE 118-Bus system, South-Carolina (SC) synthetic grid 500-Bus system, and Polish 2383-Bus system. Moreover, model reduction of a stochastic SCUC (SSCUC) was demonstrated utilizing a modified IEEE 24-Bus system with renewable generation. Simulation results demonstrate a high training accuracy to identify commitment schedule while FL and post-process ensure ML predictions do not lead to infeasible solutions with minimal loss in solution quality.

PARKINSON’S DISEASE PREDICTION USING MACHINE LEARNING

Parkinson's disease (PD) is a condition that affects a substantial number of individuals worldwide, leading to impairments in motor functions and a decline in overall quality of life. The timely identification of PD is vital for effective intervention and better patient outcomes. This study investigates the Utilization of machine learning methods in predicting and diagnosing Parkinson's disease using clinical and biomedical data.The dataset is pre-processed to address missing values and standardize features. Subsequently, various ML algorithms, including Support Vector Machines (SVM) are employed to develop predictive models.The research evaluates the performance of each model through rigorous training and testing procedures, utilizing metrics such as accuracy, sensitivity, and specificity. Feature importance analysis is conducted to identify key factors contributing to the predictive accuracy of the models. Additionally, the study investigates the impact of different feature subsets on model performance.The proposed ML models exhibit promising results in accurately predicting Parkinson's disease based on diverse sets of features. The research contributes to the ongoing efforts to develop non-invasive and efficient diagnostic tools for Parkinson's disease, providing a foundation for further studies and potential integration into clinical practice. A linear kernel utilized in a support vector machine classifier trained on a dataset with diverse voice-related features, demonstrated exceptional performance in predicting Parkinson's disease. The user interface was designed with a focus on availability, make secure that users can easily navigate and understand the application. The prediction result, either An individual has been Received a diagnosis of Parkinson's disease or The person does not have Parkinson's disease," is displayed. The model achieved an accuracy of (87%) on the test setThe high training accuracy of (86%) further affirms the model's capability to capture underlying patterns in the training data. Keywords: Parkinson's disease, machine learning, predictive modelling, feature importance, neurodegenerative disorders.

Classifying Bengali Newspaper Headlines with Advanced Deep Learning Models: LSTM, Bi-LSTM, and Bi-GRU Approaches

Reading newspapers is beneficial for people of all ages and the global community. The enjoyment of gathering diverse data from various sources adds to the overall experience. To enhance specificity in Bengali news headlines, recognizing the news genre becomes crucial. Recognizing the genre of the news, it is a very challenging task in Bengali Text Classification with the help of AI. A very few research works is done on Bengali News headline classification and we have done a model to provide a solution to the addressed issue. Due to the continuous change of the structure of the news headlines, we have employed a neural network adoption connection to our methodology experiment on a mixture of primary and secondary dataset. Achieving significant results, we implemented a Bengali dataset in Multi Classification using Long-Short Term Memory (LSTM), Bi- Long-Short Term Memory (Bi-LSTM), and Bi-Gated Recurrent Unit (Bi-GRU). The dataset is established by aggregating news headlines from various Bengali news portals and websites, showcasing robust categorization performance in the end product. Six categories were employed for the classification of Bengali newspaper headlines. The Bi-LSTM Model emerged with the highest training accuracy at 97.96% and the lowest validation accuracy at 77.91%. Furthermore, it demonstrated enhanced sensitivity and specificity.

Deep learning based Glaucoma Network Classification (GNC) using retinal images

AbstractThe proposed deep learning framework for glaucoma classification addresses critical challenges of limited data and computational costs. Employing data augmentation and normalization techniques, the three‐stage model, utilizing InceptionV3 and ResNet50, achieves high training (99.3% ‐ 99.8%) and testing accuracy (91.6% ‐ 92.12%) on a dataset comprising 16,328 images from fused public datasets. This outperforms existing automated models. The approach leverages transfer learning and convolutional neural networks, showcasing its potential for accurate and timely glaucoma diagnosis. However, ongoing validation on diverse datasets and ethical considerations regarding fairness and transparency in medical applications remain essential. The model's reliability suggests its promising role in aiding early glaucoma detection, potentially averting irreversible vision impairment.

The Application of Immersive Virtual Reality and Machine Learning for the Assessment of Unilateral Spatial Neglect

Abstract Even subtle forms of hemispatial neglect after stroke negatively affect the performance of daily life tasks, increase the risk of injury, and are associated with poor rehabilitation outcomes. Conventional paper-and-pencil tests, however, often underestimate the symptoms. We aimed to identify relevant neglect-specific measures and clinical decision rules based on machine learning techniques on behavioral data generated in a new Virtual Reality (VR) application, the immersive virtual road-crossing task. In total, 59 participants were included in our study: two right-hemispheric stroke groups with left neglect (N = 20) or no neglect (N = 19), classified based on conventional tests and medical diagnosis, and healthy controls (N = 20). A neuropsychological test battery and the VR task were administered to all participants. We applied decision trees and random forest models to predict the respective groups based on the results of the VR task. Our feature selection procedure yielded six features as suitable predictors, most of which involved lateral time-related measures, particularly reaction times, and head movements. Our model achieved a high training accuracy of 96.6% and estimated test accuracy of 76.8%. These results confirm previous reports that temporal behavioral patterns are key to detecting subtle neglect in patients with chronic stroke. Our results indicate that VR combined with machine learning has the potential to achieve higher test accuracies while being highly applicable to clinical practice.