The sine cosine algorithm (SCA) is a well-known stochastic optimizer that uses the oscillation characteristic and periodicity of the sine–cosine mathematical model to construct operators to accomplish the search objective iteratively with the merit of a simple structure and easy to understand. However, SCA possesses several limitations, such as low convergence velocity and accuracy, as well as trapping into local stagnation. To overcome these difficulties, this paper combines three mechanisms, namely greedy selection (GS), dimension learning-based hunting (DLH), and terminal replacement mechanism (TRM), to propose a boosted SCA, named GDTSCA. In this research, the GS strategy plays a role in reducing the implementation of invalid position updates by individuals and facilitating individuals to search in a better direction, which enhances the convergence rate and precision of GDTSCA. The DLH strategy uses different methods to build a neighborhood for each searching individual to accomplish information sharing among the individuals. The strategy can mitigate the problem of diversity reduction of searching individuals and boost the global search. The TRM strategy takes the neighbors of the best individual to update the worst individual to keep the individual from falling into local stagnation. To scientifically analyze and verify the performance of GDTSCA, this paper conducts experiments based on the CEC2017 test suite. First, the proposed algorithm is analyzed for the impact of different strategies. Then, this paper studies the search history and balance diversity of GDTSCA utilizing qualitative analysis. Next, to evaluate the ability of GDTSCA, it is compared with nine original meta-heuristics, seven SCA variants, and six other advanced algorithms. The results show that GDTSCA exhibits a stronger optimization effect than the optimizers utilized in most benchmark functions. Finally, the performance of GDTSCA for optimization in real scenarios is explored based on three well-known engineering design tasks. The outcomes show that the optimization ability of GDTSCA is superior to that of many well-known algorithms utilized in the experiment.

Lupus nephritis (LN), a severe kidney complication of systemic lupus erythematosus, necessitates prompt diagnosis and treatment to prevent kidney failure and other serious health consequences. To assist more accurately in the pathological diagnosis of LN, researchers have explored an innovative method that combines multi-threshold image segmentation and a metaheuristic algorithm for LN classification. However, it is worth noting that the traditional multi-threshold image segmentation technique based on a metaheuristic algorithm often faces the dilemma of falling into local optimal solutions during the image segmentation process, making determining the optimal threshold set quite difficult. To address this issue, we propose the AFSRIME, a variant of the rime optimization algorithm (RIME). Specifically, we modify the rime factor of RIME and introduce an adaptive rime factor, replacing the optimal global position with the optimal position of each agent. This modification aims to alleviate the convergence stagnation problem encountered by RIME in its later updates and mitigate its propensity to fall into local optima. We further incorporated a foraging and storage strategy to enhance information sharing within the RIME populations and fully exploit the intra-group information. To verify the performance of AFSRIME, we conducted a series of tests using the IEEE CEC 2017 benchmark functions and compared them with multiple cutting-edge algorithms. The validation results indicated AFSRIME's impressive performance. Finally, we applied AFSRIME to multi-threshold segmentation of LN images using Renyi's entropy non-local means two-dimensional histogram segmentation technique. The results show that AFSRIME enhanced RIME's multi-threshold segmentation capability on LN images, demonstrating a distinct advantage compared to other advanced algorithms. These findings underscore AFSRIME's promising future in LN image segmentation applications.

In this paper, we propose an improved SMA algorithm (SRSMA), incorporating a Sobel sequence mechanism and Rosenbrock mechanism. The next task involves constructing a machine learning model called SRSMA-FKNN to predict tuberculous pleural effusion by combining SRSMA with Fuzzy k-nearest neighbor (FKNN). To verify the capability of the new algorithm, using 30 IEEE CEC2017 competition functions, comparative experiments are carried out on nine classical meta-heuristic algorithms, eleven improved algorithms, and SRSMA. The findings of the experiments with the mean, standard deviation, Friedman test, and Wilcoxon signed rank test demonstrated that SRSMA was the most successful method overally. To further validate the utility of our proposed SRSMA-FKNN framework, we conducted experiments using clinical data collected from patients with suspected tuberculous pleural effusion. Our results indicate that key indicators, such as pleural effusion adenosine deaminase (PEADA), glucose (GLU), urea nitrogen/creatinine ratio (BUN/CR), lactate dehydrogenase (LDH), and pleural effusion lymphocytes percentage (PELP) are critical for the feature selection suggested in this study to determine the severity of tuberculous pleural effusion. According to the classification results, the prediction model demonstrates an accuracy of 88.57

Feature selection (FS) is a critical preprocessing step in machine learning to enhance model performance and interpretability. However, the increasing dimensionality of data poses significant challenges for traditional metaheuristic algorithms (MAs), including the novel moss growth optimizer (MGO), which often struggle to balance exploration and exploitation and handle high-dimensional datasets effectively. To address these limitations, we propose QMGO, an MGO enhanced by reinforcement learning (RL) decision-making. QMGO incorporates the random following (RF) strategy to improve exploration and the weighted mean of vectors (INFO) strategy to boost exploitation. A Q-learning mechanism is employed to dynamically select the optimal strategy between RF and INFO during different optimization stages. Furthermore, a binary version, bQMGO, is developed specifically for FS tasks. On the 29 CEC 2017 benchmark functions, QMGO demonstrated exceptional optimization capability, achieving the first rank against both 11 original and 11 advanced algorithms. In feature selection experiments across 14 public datasets, bQMGO achieved zero classification error on all datasets while selecting the minimum number of features. These results confirm that QMGO, by integrating a RL decision framework, is a powerful and reliable tool for feature selection, enriching the application of metaheuristic algorithms in this domain.

Alzheimer's Disease (AD) is a progressive neurodegenerative disorder marked by neuronal loss, leading to cognitive and behavioral decline. With the aging global population, AD incidence and its socioeconomic burden are increasing. Developing effective early diagnostic methods is thus critical for improving patient outcomes and slowing disease progression. In this paper, an enhanced Particle Swarm Optimization (PSO) algorithm, which integrates opposition-based Latin squares sampling initialization (OL) with dynamic inertia weights and learning factors (D), termed OLDPSO, is proposed to improve feature selection and classification within a Support Vector Machine (SVM) model for AD diagnosis using magnetic resonance imaging (MRI) data. MRI, as a non-invasive modality, reveals structural brain changes, particularly in gray matter (GM) and white matter (WM) volumes, which are key biomarkers for AD. However, extracting essential features from complex GM and WM data remains a significant challenge. To address this, the proposed OLDPSO, which adaptively balances global exploration and local exploitation, overcomes traditional PSO limitations. Benchmark experiments show that OLDPSO outperforms existing PSO variants in solution quality and convergence speed. Validated with data from the AD Neuroimaging Initiative (ADNI), the OLDPSO-SVM model demonstrates superior performance in differentiating AD, mild cognitive impairment (MCI), and normal control (NC) groups, particularly in classifying MCI subtypes (MCI-NC and MCI-C). Results show that combining GM and WM features yields higher diagnostic accuracy than using either alone, and the model identified key brain regions associated with AD progression. Specifically, the model achieved accuracies of 99.11%, 89.76%, 99.07%, 88.38%, 94.69%, and 87.96% in the diagnosis of AD vs. NC, NC vs. MCI-NC, NC vs. MCI-C, MCI-NC vs. MCI-C, MCI-NC vs. AD, and MCI-C vs. AD, respectively. Through optimized feature selection, the OLDPSO-SVM model enhances diagnostic performance and provides valuable insights for developing MRI-based multimodal diagnostic tools for AD.

Surrogate-assisted evolutionary algorithms are widely used for solving expensive and complex optimization problems, yet existing methods face persistent challenges in balancing accuracy and efficiency, particularly in high-dimensional spaces. This paper presents SRIME, an advanced surrogate-assisted RIME algorithm that introduces three key innovations to overcome these limitations. First, our elite sample sampling mechanism generates high-quality initial populations while maintaining computational efficiency. Second, we develop an improved Gaussian regression model that dynamically incorporates historical optimal data during optimization for more accurate function approximation. Most significantly, we propose two novel strategies with theoretical guarantees: (1) an adaptive long-distance dynamic search that automatically adjusts exploration ranges based on population distribution to prevent premature convergence, and (2) a mutation-assisted local search that provides efficient refinement in promising regions. Comprehensive theoretical analysis demonstrates SRIME's convergence properties and computational complexity advantages, Statistical tests confirm these improvements are significant (p < 0.01). We further validate SRIME's practical utility through two important applications. In photovoltaic parameter estimation, SRIME achieves highly accuracy under challenging outdoor conditions, showing particular robustness to temperature and light variations. These results demonstrate SRIME's effectiveness in both synthetic benchmarks and real-world optimization problems, with special strengths in handling high-dimensional scenarios that challenge existing methods.

BACKGROUND:Brachytherapy planning places radioactive sources around the tumor target to ensure that the tumor is covered by the prescribed dose while maximizing dose sparing for normal organs. In brachytherapy, the mainstream inverse planning methods are Hybrid Inverse Planning Optimization and Inverse Planning Simulated Annealing. Most existing algorithms use a single objective function with weighting parameters to balance tumor and normal tissue importance, but choosing these weights is often subjective and difficult, which can result in biased or suboptimal plans. PURPOSE:To propose a seed implantation position optimization algorithm in prostate low-dose-rate brachytherapy (LDRBT) to improve dose distribution and meet clinical dosimetric criteria. METHODS:A multi-objective non-dominated genetic algorithm was proposed to optimize seed implantation positions. The optimization focused on direct optimization of dose-volume indices and incorporated inverse planning to adjust implantation control parameters. A prostate case with four clinical dose constraints was tested. The proposed algorithm was compared against single-objective optimization to evaluate its effectiveness in handling discrete variables and managing multiple constraints as well. RESULTS:Single-objective optimization showed limitations, with only 10% of plans meeting some conditions and none satisfying all four constraints. In contrast, the multi-objective optimization approach produced a larger set of Pareto front solutions, with 24.8% of plans meeting some conditions and six plans fully satisfying all constraints. Although the maximum urethral dose was higher in the multi-objective optimization method, it successfully balanced all constraints. CONCLUSIONS:Multi-objective optimization offers significant advantages over single-objective methods in prostate LDRBT, providing a more comprehensive set of feasible solutions that balance conflicting goals. It shows better performance in achieving plans that meet all clinical constraints.

Alzheimer's disease is an irreversible central neurodegenerative disease, and early diagnosis of Alzheimer's disease is beneficial for its prevention and early intervention treatment. In this study, we propose a novel framework, FusionNet-ISBOA-MK-SVM, which integrates a fusion network (FusionNet) and improved secretary bird optimization algorithm to optimize multikernel support vector machine for Alzheimer's disease diagnosis. The model leverages multimodality data, including functional magnetic resonance imaging and genetic information (single-nucleotide polymorphisms). Specifically, FusionNet employs U-shaped hierarchical graph convolutional networks and sparse graph attention networks to select feature effectively. Extensive validation using the Alzheimer's Disease Neuroimaging Initiative dataset demonstrates the model's superior interpretability and classification performance. Compared to other state-of-the-art machine learning methods, FusionNet-ISBOA-MK-SVM achieves classification accuracies of 98.6%, 95.7%, 93.0%, 91.8%, 93.1%, and 95.4% for HC vs. AD, EMCI vs. AD, LMCI vs. AD, EMCI vs. AD, HC vs. EMCI, and HC vs. LMCI, respectively. Moreover, the proposed model identifies affected brain regions and pathogenic genes, offering deeper insights into the mechanisms and progression of Alzheimer's disease. These findings provide valuable scientific evidence to support early diagnosis and preventive strategies for Alzheimer's disease.

Lack of exploration capability is an important constraint on the performance of multi-agent deep reinforcement learning (MADRL) algorithms. Through evolutionary selection, humans can quickly explore unfamiliar environments and master survival skills by using the episodic memory in the brain. Inspired by biology, this paper proposes a multi-agent reinforcement learning exploration framework based on causal episodic memory and potential evolution (MACMPE) to improve the exploration capability of multi-agent systems in different types of task scenarios. First, we construct a causal episodic memory module that introduces causal learning computation and selects samples with high causal impact for the current training phase for policy updating to accelerate the policy learning process. Next, by combining the advantages of breadth search of evolutionary algorithms and deep search of deep learning, we build a potential evolution (PE) module to improve the ability of the multi-agent system to find the optimal solution in the environment. Then, we combine and embed the CM module and the PE module into the MADRL algorithm MAAC. Finally, we experiment with task scenarios including cooperative collection, command movement, and target navigation, and extend this framework to different MADRL algorithms. Experimental results show that the MADRL algorithms, combined with the framework proposed in this study, outperform the baseline algorithm regarding exploration capability and have better universality for the number of agents and scene categories.

As the global population ages, Alzheimer’s disease (AD) poses a significant worldwide challenge as a leading cause of dementia, with a slow early progression that eventually leads to nerve cell death and currently lacks effective treatment. However, early diagnosis can slow its progression through pharmaceutical intervention, making accurate early diagnosis using computer-aided diagnosis (CAD) systems crucial. This study aims to enhance the accuracy of early AD diagnosis by developing an improved optimization approach for deep learning-based CAD systems. To achieve this, this paper proposes an improved Harris Hawks optimization algorithm (HHO), named CAHHO, which incorporates crisscross search and adaptive β-Hill climbing mechanisms, thereby enhancing population diversity and search space coverage during the exploration phase, while adaptively adjusting the step size during the exploitation phase to improve local search precision. Comparative experiments with classical algorithms, HHO variants, and advanced optimization methods validate the superiority of the proposed CAHHO. Specifically, this study employs the deep learning model residual network with 18 layers (ResNet18) as the base model for AD diagnosis and uses CAHHO to optimize key hyperparameters, including the number of channels and learning rate. Experiments on the AD neuroimaging initiative dataset demonstrate that the ResNet18-CAHHO model outperforms existing methods in classifying AD, mild cognitive impairment (MCI), and normal control (NC) subjects. Specifically, it achieves accuracies of 0.93077, 0.80102, and 0.80513 in the diagnosis of AD versus NC, MCI versus NC, and AD versus MCI, respectively. Furthermore, Gradient-Weighted Class Activation Mapping (Grad-CAM) visualizations reveal critical brain regions associated with AD, providing valuable diagnostic support for clinicians and holding significant promise for early intervention.

Reconstructing magnetic resonance (MR) images from undersampled k-space data have always been a challenging problem. Compressed Sensing (CS) can reconstruct images from a small amount of sampled data when combined with the robust feature learning ability of deep learning, and it can further reduce the sampling time. Most previous deep learning methods using compressed sensing relied heavily on convolutional neural networks(CNNs) or swin transformer block(STB), even they reconstructed images through stacking or cross-domain structure. However, due to the limited size of their receptive fields, convolutional neural networks cannot explore the global features of images. Conversely, vast receptive fields would increase model complexity and make the entire network difficult to train. In this paper, we proposed a cascade dual-domain swin-conv unet for reconstruction(CDSCU-Net), which combines STB and CNNs to focus on both local and global features during reconstruction. By fusing these features through incorporating our designed residual modules in the skip connections, we can mine more refined feature representations. Compared with the best-performing deep learning reconstruction methods based on compressed sensing in recent years, CDSCU-Net can better preserve the structural details of images and achieve good reconstruction quality at lower acceleration factors, additionally, the reconstructed images can also serve as raw data for other tasks.

This study presents an innovative approach to enhancing the performance of perovskite solar cells through the integration of a functionally graded triply periodic minimal surface (FG-TPMS) layer. The research focuses on the mechanical and vibrational characteristics of doubly curved panels embedded with three distinct iterations of the FG-TPMS model: the primitive, gyroid, and wrapped package graph (IWP). By employing higher-order shear deformation theory (HSDT), the analysis accounts for the complex geometrical and material gradations within the FG-TPMS structures. An advanced analytical method utilizing trigonometric functions is developed to accurately predict the natural frequencies and mode shapes of these novel composite structures. In order to assess the vibrations of TPMS-reinforced perovskite solar cells surrounded by an elastic foundation, this work proposes the implementation of a novel Support Vector Machine (SVM)-deep neural network (DNN)-Genetic Algorithm (GA) employing mathematical modeling datasets. Using the SVM-DNN-GA algorithm, predicted accuracy is improved. In order to simulate and forecast the vibrational behavior of the reinforced solar cells, the integrated methodology makes use of the advantages of each technique. The results indicate that the integration of FG-TPMS layers significantly enhances the mechanical stability of the perovskite solar cells. The application of HSDT reveals detailed insights into the dynamic responses of the doubly curved panels, highlighting the potential for fine-tuning their vibrational characteristics to further improve solar cell performance. This research underscores the potential of FG-TPMS structures in advancing solar cell technology, providing a foundation for future studies to explore the integration of complex geometries and material gradations in photovoltaic applications.

As a subfield of deep learning (DL), generative adversarial networks (GANs) have produced impressive generative results by applying deep generative models to create synthetic data and by performing an adversarial training process. Nevertheless, numerous issues related to the instability of training need to be urgently addressed. Evolutionary computation (EC), using the corresponding paradigm of biological evolution, overcomes these problems and improves evolutionary-based GANs’ ability to deal with real-world applications. Therefore, this paper presents a systematic literature survey combining EC and GANs. First, the basic theories of GANs and EC are analyzed and summarized. Second, to provide readers with a comprehensive view, this paper outlines the recent advances in combining EC and GANs after detailed classification and introduces each of them. These classifications include evolutionary GANs and their variants, GANs with evolutionary strategies and differential evolution, GANs combined with neuroevolution, evolutionary GANs related to different optimization problems, and applications of evolutionary GANs. Detailed information on the evaluation metrics, network structures, and comparisons of these models is presented in several tables. Finally, future directions and possible perspectives for further development are discussed.

In recent years, brain imaging genomics has advanced significantly in revealing underlying pathological mechanisms of Alzheimer's disease (AD) and providing early diagnosis. In this paper, we present a framework for diagnosing AD that integrates magnetic resonance imaging (fMRI) genetic preprocessing, feature selection, and a support vector machine (SVM) model. In particular, a novel sand cat swarm optimization (SCSO) algorithm, named SS-SCSO, which integrates the spiral search strategy and alert mechanism from the sparrow search algorithm, is proposed to optimize the SVM parameters. The optimization efficacy of the SS-SCSO algorithm is evaluated using CEC2017 benchmark functions, with results compared with other metaheuristic algorithms (MAs). The proposed SS-SCSO-SVM framework has been effectively employed to classify different stages of cognitive impairment in Alzheimer's Disease using imaging genetic datasets from the Alzheimer's Disease Neuroimaging Initiative. It has demonstrated excellent classification accuracies for four typical cases, including AD, early mild cognitive impairment, late mild cognitive impairment, and healthy control. Furthermore, experiment results indicate that the SS-SCSO-SVM algorithm has a stronger exploration capability for diagnosing AD compared to other well-established MAs and machine learning techniques.

Alzheimer's disease (AD) diagnosis utilizing single modality neuroimaging data has limitations. Multimodal fusion of complementary biomarkers may improve diagnostic performance. This study proposes a multimodal machine learning framework integrating magnetic resonance imaging (MRI), positron emission tomography (PET) and cerebrospinal fluid (CSF) assays for enhanced AD characterization. The model incorporates a hybrid algorithm combining enhanced Harris Hawks Optimization (HHO) algorithm referred to as ILHHO, with Kernel Extreme Learning Machine (KELM) classifier for simultaneous feature selection and classification. ILHHO enhances HHO's search efficiency by integrating iterative mapping (IM) to improve population diversity and local escaping operator (LEO) to balance exploration-exploitation. Comparative analysis with other improved HHO algorithms, classic meta-heuristic algorithms (MHAs), and state-of-the-art MHAs on IEEE CEC2014 benchmark functions indicates that ILHHO achieves superior optimization performance compared to other comparative algorithms. The synergistic ILHHO-KELM model is evaluated on 202 AD Neuroimaging Initiative (ADNI) subjects. Results demonstrate superior multimodal classification accuracy over single modalities, validating the importance of fusing heterogeneous biomarkers. MRI + PET + CSF achieves 99.2 % accuracy for AD vs. normal control (NC), outperforming conventional and proposed methods. Discriminative feature analysis provides further insights into differential AD-related neurodegeneration patterns detected by MRI and PET. The differential PET and MRI features demonstrate how the two modalities provide complementary biomarkers. The neuroanatomical relevance of selected features supports ILHHO-KELM's potential for extracting sensitive AD imaging signatures. Overall, the study showcases the advantages of capitalizing on complementary multimodal data through advanced feature learning techniques for improving AD diagnosis.

This study presents an innovative approach to mitigate vibrations induced by external shock on composite structures through the application of an intelligent controller. Leveraging the first-order shear deformation panel theory, a sophisticated controller scheme is developed, integrating methodologies such as the differential quadrature approach and Laplace transform. Furthermore, deep neural network (DNN) and support vector regression (SVR) techniques are employed to enhance prediction accuracy and control efficiency. Additionally, two optimized hybrid models are proposed, incorporating Particle Swarm Optimization (PSO) and Grey Wolf Optimizer (GWO) algorithms, to further refine the controller's performance. The proposed methodology aims to address the challenges associated with vibrations in composite structures by providing a comprehensive and adaptive control solution. By utilizing advanced optimization algorithms and machine learning techniques, the controller can effectively adapt to dynamic changes in external shock conditions, thereby minimizing vibrations and ensuring structural integrity. The integration of ANN and SVR enhances the controller's predictive capabilities, enabling it to anticipate and respond to varying shock scenarios with precision. Through theoretical analysis and numerical simulations, the effectiveness of the proposed intelligent controller is demonstrated in reducing vibrations and enhancing the structural stability of composite systems. The optimized hybrid models, employing PSO and GWO algorithms, further improve the controller's performance by fine-tuning its parameters for optimal control efficiency. Overall, this research contributes to the development of robust control strategies for mitigating vibrations in composite structures subjected to external shock, with potential applications in aerospace, automotive, and civil engineering industries.

The increasing prevalence of electric vehicles underscores the need for enhanced battery pack safety, particularly against impacts that can lead to thermal runaway and fires. This study investigates the mechanical and thermal characteristics of battery packs subjected to cone impact on the lower part. Utilizing finite element model (FEM) simulations, we examined the effects of top radius, shell condition, velocity, and angle on the battery pack's response. Five machine learning (ML) techniques were employed to predict the battery pack's behavior under impact, with training data generated from a well-planned Latin hypercube experiment based on FEM dynamic simulations. The accuracy and robustness of the ML models were evaluated under various scenarios, including the introduction of Gaussian noise. Among the models tested, BESA-ELMM (Bald eagle search algorithm-Extreme learning machine model) demonstrated exceptional speed, making it suitable for real-time assessments, while WOA-SVMM (whale optimization algorithm-Support vector machine model) exhibited superior resilience and accuracy, particularly under noisy conditions. Both models, along with the other ML techniques, showed significant effectiveness in predicting the mechanical responses of battery packs to impact. Our findings indicate that ML approaches are highly efficient in evaluating the mechanical effects on battery packs, providing crucial insights for designing safer and more durable battery packs. This study contributes to the advancement of battery safety by demonstrating the potential of integrating ML techniques with FEM simulations to enhance the resilience and impact resistance of battery packs in electric vehicles.

The pathogenesis of Alzheimer's disease (AD) remains unclear, but revealing individual differences in functional connectivity (FC) may provide insights and improve diagnostic precision. A hierarchical clustering-based autoencoder with functional connectivity was proposed to categorize 82 AD patients from the Alzheimer's Disease Neuroimaging Initiative. Compared to directly performing clustering, using an autoencoder to reduce the dimensionality of the matrix can effectively eliminate noise and redundant information in the data, extract key features, and optimize clustering performance. Subsequently, subtype differences in clinical and graph theoretical metrics were assessed. Results indicate a significant inter-subject heterogeneity in the degree of FC disruption among AD patients. We have identified two neurophysiological subtypes: subtype I exhibits widespread functional impairment across the entire brain, while subtype II shows mild impairment in the Limbic System region. What is worth noting is that we also observed significant differences between subtypes in terms of neurocognitive assessment scores associations with network functionality, and graph theory metrics. Our method can accurately identify different functional disruptions in subtypes of AD, facilitating personalized treatment and early diagnosis, ultimately improving patient outcomes.

Advanced threat detection systems are needed more than ever as cyber-attacks become more advanced. A novel Neural Networks, and dense network (BFRGD-Net) architectures to identify threats with unmatched accuracy and speed. The approach optimizes threat detection using Dynamic Range Realignment, anomaly-driven feature enhancement, and a hybrid feature selection strategy on a comprehensive Texas dataset of 66 months of real-world network activity. With 97.8% accuracy, 97.5% F1-score, and 98.3% AUC, BFRGD-Net sets new standards in the field. Threat Detection Sensitivity shows the model's capacity to find uncommon, high-severity threats, while Balanced Risk Detection Efficiency provides fast, accurate threat detection. The model has strong correlations and the highest statistical metrics scores compared to other techniques. Extensive simulations demonstrate the model's capacity to discern threat levels, attack kinds, and response techniques. BFRGDNet revolutionizes cybersecurity by seamlessly merging cuttingedge machine learning with specific insights. Its advanced threat detection and classification engine reduces false negatives and enables proactive critical infrastructure protection in real-time. The model's adaptability to various attack situations makes it vital for cybersecurity resilience in a digital environment.