Heart disease remains a leading cause of mortality worldwide, emphasizing the urgent need for reliable and interpretable predictive models to support early diagnosis and timely intervention. However, existing Deep Learning (DL) approaches often face several limitations, including inefficient feature extraction, class imbalance, suboptimal classification performance, and limited interpretability, which collectively hinder their deployment in clinical settings. To address these challenges, we propose a novel DL framework for heart disease prediction that integrates a comprehensive preprocessing pipeline with an advanced classification architecture. The preprocessing stage involves label encoding and feature scaling. To address the issue of class imbalance inherent in the personal key indicators of the heart disease dataset, the localized random affine shadowsampling technique is employed, which enhances minority class representation while minimizing overfitting. At the core of the framework lies the Deep Residual Network (DeepResNet), which employs hierarchical residual transformations to facilitate efficient feature extraction and capture complex, non-linear relationships in the data. Experimental results demonstrate that the proposed model significantly outperforms existing techniques, achieving improvements of 3.26% in accuracy, 3.16% in area under the receiver operating characteristics, 1.09% in recall, and 1.07% in F1-score. Furthermore, robustness is validated using 10-fold cross-validation, confirming the model’s generalizability across diverse data distributions. Moreover, model interpretability is ensured through the integration of Shapley additive explanations and local interpretable model-agnostic explanations, offering valuable insights into the contribution of individual features to model predictions. Overall, the proposed DL framework presents a robust, interpretable, and clinically applicable solution for heart disease prediction.

Internet of Things (IoT) environments are difficult to detect intrusion because of the heterogeneity of the devices and intricate attack patterns. Class imbalance, irrelevant features, suboptimal accuracy, poor hyperparameter tuning, and limited labeling data are some of the problems that exist in traditional machine and deep learning models. To address these challenges, we propose the Deep Optimized Active Learning Framework for Intrusion Detection in IoT Systems (DOAL-IDS). First, low-variance features are removed using a variance threshold, and the proximity weighted random affine shadow sampling technique balances the dataset by generating synthetic samples near the minority class. Capsule Network (CapsNet) is employed for classification and further optimized using the Reptile Search Algorithm (RSA), forming Reptile-Optimized Capsule Network (ROC-Net). ROC-Net is enhanced with Margin-Based Active Learning (MBAL) to create MARCO-Net, which efficiently annotates the most uncertain samples, reducing labeling costs. Experimental results show that ROC-Net and MARCO-Net achieve improvements of 8.75% and 12.5% in accuracy, 7.41% and 11.11% in F1-score, and 11.94% and 20.90% in Matthews correlation coefficient, with reductions of 75.86% and 82.70% in log loss. Robustness is ensured via 10-fold cross-validation and the Wilcoxon signed-rank test. For interpretability, local interpretable model-agnostic explanations and Shapley additive explanations analyzes provide insights into model decisions. These results demonstrate the effectiveness, reliability, and transparency of the proposed DOAL-IDS framework for IoT intrusion detection.

The advent of the Internet of Things (IoT) enables billions of devices in wide-ranging domains such as healthcare, industry, and smart cities to interconnect with each other, but these connections make the network vulnerable to advanced cyber threats too. Current intrusion detection methods have failed to provide effective detection capabilities mainly because of issues such as extremely imbalanced data distributions, low classification accuracy, or static and manually tuned hyperparameters that do not generalize well in dynamic IoT settings. These challenges are exacerbated by unique IoT constraints, including limited device resources and dynamic attack patterns, which further complicate effective detection. To address these challenges, in this study we present a Reinforcement-enhanced Generative Artificial Intelligence (ReGAIN) framework for intelligent intrusion detection in IoT networks. In this approach, we use a generative autoencoder for data balancing to generate realistic minority class instances in the latent feature space, and meanwhile to obtain stable and unbiased learning of the model. This paper introduces a novel Pointer-Attention Dual Network (PAD-Net) that employs a Dual Attention Network (DANet) and a Pointer Network (PtrNet) to enhance spatial attention and inter-feature relationships. We also propose Reinforcement-enhanced PAD-Net (RePAD-Net), which leverages reinforcement learning to automatically optimize key hyperparameters at each training step, further enhancing generalization ability and robustness. The intrusion detection task in this study is a multi-class classification problem, where different types of attacks are distinguished from each other. Experimental results demonstrate that PAD-Net and RePAD-Net achieve notable improvements of 3.79

Cryptocurrency price prediction poses significant challenges due to the inherent volatility and nonlineardynamics of the market. This study introduces a hybrid stacked modeling framework that integrates machine learning (ML) and deep learning (DL) techniques, capitalizing on their complementary strengths-ML models are effective at capturing nonlinearfeature interactions in structured data, while DL architectures are adept at modeling temporal dependencies in sequential data. The proposed model leverages historical price data, technical indicators, macroeconomic variables, and sentiment metrics, with feature engineering applied to enhance predictive capability. Empirical evaluation was conducted through two experimental setups: (i) short-term, monthly segment analysis and (ii) long-term generalization via five-fold cross-validation. The hybrid model outperformed individual baseline models, achieving up to 18.3 % lower RMSE and 6.7 % higher directional accuracy. Additionally, it yielded superior risk-adjusted returns, with Sharpe Ratios reaching 0.094 on the Ethereum dataset. Beyond technical improvements, this research offers foresight into digital financial markets, providing a robust tool for investors, institutions, and policymakers navigating the evolving cryptocurrency landscape. The model supports more informed decision-making, enhances market oversight, and contributes to the development of adaptive regulatory frameworks for digital finance.

Intrusion detection systems play major role in security of Internet of Things (IoT) networks against various types of cyber threats. However, traditional Machine Learning (ML) and Deep Learning (DL) models still struggle with challenges such as critical class imbalance, reliance on manual tuning of hyperparameters, and the expensive cost of obtaining labeled data. In this study, we addressed these limitations using the TON_IoT dataset. To address data imbalance, the proximity weighted random affine shadow sampling technique is utilized. Further, Bayesian optimization is applied on LeNet, resulting in LeBayesNet, which provides the optimal configuration for high-accuracy threat detection. Additionally, EntroLeNet integrates entropy-based uncertainty into the learning process for improved robustness. Next, MargiLeNet leverages marginal-based active learning, annotating the most uncertain samples. Experimental results obtained show that LeBayesNet, MargiLeNet, and EntroLeNet improve performance over existing ML and DL models by 6.90%, 5.80%, and 4.27% in accuracy and 6.19%, 6.43%, 8.05%, and 7.52% in receiver operating characteristic-area under the curve, respectively. The LeBayesNet, MargiLeNet, and EntroLeNet models significantly reduce Hamming loss by 72.84%, 60.49%, and 43.21%, respectively. For robustness and generalizability assurance, 10-fold cross-validation is used during evaluation and validated the statistical significance of performance improvements using the Mann–Whitney U test. Furthermore, the models’ explainability and interpretability is enhanced through Shapley additive explanations and local interpretable model-agnostic explanations, providing insights into feature importance and decision transparency.

Intrusion Detection (ID) in the Internet of Secure Things (IoST) has become increasingly critical due to the rising frequency and sophistication of cyber-attacks, which can lead to severe consequences such as data breaches, financial losses, and service disruptions. These risks are further intensified in computationally limited environments, where limited computational capacity and rapidly evolving threats make accurate and efficient detection challenging. In this study, a data-efficient ID framework tailored for resource-constrained environments is proposed by leveraging active learning and meta-heuristic optimization techniques. The proposed framework systematically addresses three critical limitations commonly observed in traditional models: data imbalance, inefficient hyperparameter tuning, and dependency on large labeled datasets. Initially, to mitigate class imbalance, adaptive synthetic sampling generates synthetic instances for minority classes, thereby enhancing learning in complex regions of the feature space. Next, for hyperparameter optimization, the Sandpiper Optimization (SO) algorithm fine-tunes the regularization parameter of Logistic Regression (LR), yielding significant improvements in model generalization. Finally, the challenge of limited labeled data is addressed through two active learning strategies: Active Learning Uncertainty-based (ALU) and Active Learning Entropy-based (ALE). These strategies selectively query the most informative samples from the unlabeled pool, ensuring maximum learning with minimal annotation effort. The performance of the proposed models is evaluated on two benchmark datasets: the wireless sensor networks and network intrusion detection datasets. Simulation results demonstrate that proposed models outperform base model LR. LRALE achieves improvements of 10.48% and 3.16% in accuracy, 19.48% and 3.16% in recall, and 7.23% and 1.04% in F1-score on WSN-DS and CIC-IDS-DS datasets, respectively. LRALU shows improvements of 18.18% and 2.11% in accuracy, 18.18% and 2.11% in recall, and 14.63% and 2.08% in Receiver Operating Characteristic-Area Under the Curve (ROC-AUC). Similarly, LRSO achieves improvements of 9.09% and 2.11% in accuracy, 9.09% and 1.05% in recall, and 9.76% and 3.12% in ROC-AUC on WSN-DS and CIC-IDS-DS datasets, respectively. To ensure model generalization and stability across different data partitions, a rigorous 10-fold cross-validation is conducted. Model interpretability is further enhanced using eXplainable artificial intelligence techniques, including Local interpretable model-agnostic explanations and Shapley additive explanations, to elucidate feature contributions and improve transparency. Additionally, statistical significance testing through paired t-tests confirms the robustness and reliability of the proposed models. Overall, this framework introduces a comprehensive, annotation-efficient, and transparent ID solution that significantly advances the domain, making it well-suited for practical deployment in IoSTs environments.

A powerful multi-stage framework is proposed, which includes dimensionality reduction, adaptive sampling, attention-driven learning, and smart query selection. As the first, Least Absolute Shrinkage and Selection Operator (LASSO) is used to maintain the predictive attributes as much as possible and as the second, Proxy Weighted Random Affine Shadow Sampling (ProWRAS) is used to balance skewed classes artificially. An attention network is then used to focus on meaningful input dimensions in the classification Feature Attention Network (FAN). This system is also improved through the addition of uncertainty-based active learning which gradually extends the labeled dataset through high-entropy queries, and a Bayesian optimization cycle that automatically optimizes hyperparameters by searching through the highest performance points. Empirical results show a marked improvement over traditional methods, with accuracy gains of 39% compared to HighwayNet, 31% over GRU, and 12% over Gradient Boosting. F1-score and AUC-ROC also increased substantially, demonstrating the effectiveness of the proposed design in delivering high detection reliability. The final model offers a scalable and energy-aware solution, tailored to the unique demands of real-world WSNs.

Diabetes has become a severe and widespread chronic disease, making its early prediction crucial for effective intervention and management. With its rising global incidence, the development of robust and efficient predictive models is essential for timely diagnosis and accurate prediction. Despite the availability of numerous predictive models, existing methods often struggle with class imbalance, suboptimal and biased classification performance, and limited interpretability. To overcome these issues, this study proposes a trustworthy and patient privacy-aware framework for early and precise diabetes prediction. Firstly, the class imbalance issue in the dataset is mitigated by the proximity-weighted synthetic oversampling technique, which minimizes biased classification. Later, a novel deep learning model named Deep Residual Network (DRNet) is proposed to improve predictive performance by identifying complex patterns in the data. Experimental results demonstrate that DRNet outperforms baseline models, achieving improvements of 4.54% in accuracy, 3.48% in precision, 7.69% in Cohen's Kappa, 6.33% in Matthew's correlation coefficient, 2.28% in area under the receiver operating characteristic curve, 2.28% in precision-recall area under the curve, and 1.06% in recall. DRNet also exhibits a significant reduction in log loss with 30.43% and Hamming loss with 30%. The reliability of DRNet is confirmed using 10-fold cross validation and the Friedman test to assure consistency and robustness. Finally, explainable artificial intelligence techniques, such as local interpretable model-agnostic explanations and Shapley additive explanations, offer insights into the most contributing features both locally and globally, which leads to better interpretability and explainability. This framework aligns with the rising need for responsible AI solutions in healthcare, where transparency, trust, and data privacy have become necessary for patient-focused innovation.

The exponential proliferation of the Internet of Things (IoT) has significantly intensified the exposure of connected ecosystems to complex cyber threats, necessitating intelligent and adaptive intrusion detection systems. This study presents a comprehensive hybrid deep learning framework that unifies data balancing, feature normalization, dual-path convolutional fusion, and reinforcement-driven optimization for securing IoT infrastructures. Initially, conditional tabular generative adversarial network is utilized to alleviate severe class imbalance within the canadian institute for cybersecurity IoTs dataset, ensuring equitable representation across intrusion categories. Subsequently, min–max scaling stabilizes gradient propagation, followed by the introduction of the hybrid AS-Net architecture that fuses AlexNet’s hierarchical feature abstraction with ShuffleNet’s efficient grouped convolution through a dual-path fusion mechanism. This integration achieves superior representational diversity while maintaining computational efficiency. Furthermore, the framework is extended to a reinforcement-optimized variant, RAS-Net, wherein the monte carlo policy gradient algorithm dynamically tunes critical hyperparameters such as learning rate and dropout probability. Experimental evaluation demonstrates that AS-Net attains an accuracy of 0.8760, precision of 0.8967, recall of 0.8760, and F1-score of 0.8650, corresponding to an improvement of 3.63% in accuracy and 2.53% in F1-score over AlexNet. The reinforcement-optimized RAS-Net further enhances performance, achieving an accuracy of 0.9059, precision of 0.9267, recall of 0.9160, and F1-score of 0.9050, marking an overall increase of 6.06% in accuracy and 5.63% in F1-score compared to baseline.

In IoT networks, traffic data are high-dimensional, complex, and highly imbalanced. Minority attack classes are often underrepresented degrading the detection performance. In addition, manual labeling of large-scale IoT traffic is costly, and network behavior continuously changes over time. To overcome these challenges, this paper presents an adaptive, secure and explainable IoT intrusion detection framework. First, a generative AI based data balancing method is applied using the real-valued non-volume preserving model. It generates realistic minority-class samples while preserving the original data distribution. Then, a hybrid Deep Dual-Attention Network (DDANet) is proposed. It combines a Deep Neural Network (DNN) with a Dual Attention Network (DANet). The DNN captures global nonlinear features. The DANet focuses on important channel and spatial features. This improves feature learning and intrusion detection accuracy. Next, Active DDANet (ADDANet) integrates a monte-carlo based active learning strategy. It selects informative samples using predictive uncertainty to reduce redundant labeling and improves learning efficiency. Reinforced DDANet (RDDANet) further integrates an actor-to-critic reinforcement learning mechanism. It enables adaptive optimization in dynamic IoT environments and stabilizes training. A blockchain layer is also integrated. It ensures secure logging, decentralized trust, and tamper-resistant storage of detection results. This increases transparency and reliability in distributed IoT systems. Experimental results show that DDANet improves accuracy, recall, and F1-score by 3.33%, precision by 3.30%, Cohen's Kappa (CK) and Mathew's Correlation coefficient (MCC) by 4.60%, and Precision Recall Area Uner the Curve (PR-AUC) by 3.16% over baseline models. It also reduces Log Loss (LL) by 34.38% and Hamming Loss (HL) by 30.00%. ADDANet further improves accuracy, recall, and F1-score by 3.23%, CK and MCC by 3.30%, and reduces LL by 38.10% and HL by 42.86% over DDANet. RDDANet achieves similar accuracy improvement of 3.23%, increases CK and MCC by 4.40%, and provides the highest LL reduction of 42.86% and HL reduction of 42.86%. The framework is validated using 10-fold cross-validation to ensure robustness and generalization. Confidence interval analysis and paired t-test confirm statistical significance. Finally, integrated gradients based explainable AI is applied to provide feature attribution scores and improve model transparency. The proposed framework provides an adaptive, secure, explainable, and high-performing IoT intrusion detection solution.

Diabetes, a chronic condition caused by insufficient insulin production in the pancreas, presents significant health risks. Its increasing global prevalence necessitates the development of accurate and efficient predictive algorithms to support timely diagnosis. While recent advancements in deep learning (DL) have demonstrated potential for diabetes prediction, conventional models face limitations in handling class imbalance, capturing complex feature interactions, and providing interpretability for clinical decision-making. This paper proposes a DL framework for diabetes mellitus prediction. The framework ensures high predictive accuracy by integrating advanced preprocessing, effective class balancing, and a novel EchoceptionNet model. An analysis was conducted on a diabetes prediction dataset obtained from Kaggle, comprising nine features and 100,000 instances. The dataset is characterized by severe class imbalance, which is effectively addressed using a proximity-weighted synthetic oversampling technique, ensuring balanced class distribution. EchoceptionNet demonstrated notable performance improvements over state-of-the-art deep learning models, achieving a 4.39% increase in accuracy, 8.99% in precision, 2.19% in recall, 5.55% in F1-score, and a 7.77% in area under the curve score. Model robustness and generalizability were validated through 10-fold cross-validation, demonstrating consistent performance across diverse data splitting. To enhance clinical applicability, EchoceptionNet integrates explainable artificial intelligence techniques, Shapley additive explanations, and local interpretable model-agnostic explanations. These methods provide transparency by identifying the critical importance of features in the model’s predictions. EchoceptionNet exhibits superior predictive accuracy and ensures interpretability and reliability, making it a robust solution for accurate diabetes prediction.

A micro-grid (MG) is a localized organization of generation with a few loads, and they are becoming a more and more popular concept in the research community. With the passage of time, it is increasing the efficiency and reliability of the power systems. Both islanded mode and grid-connected mode (GCM) are used by MGs. MGs use the software application to replace their strength in order to meet the standards of GCM. However, line losses are considerable since MG and the macro station (MS) are located farther apart. Therefore, a hierarchical coordination model (HCM) is proposed for successfully replacing the power among MGs. Since some of the distances among the MGs are shorter than the distance between the MG and MS, HCM seeks to limit the power line losses through making hierarchical coalitions. In addition, a pricing scheme is proposed for encouraging the MGs to participate in the HCM. This pricing scheme is implemented along with the HCM, and both purchasing and selling costs of each MG are compared before and after coordination. Numerical simulations demonstrated that the recommended pricing scheme succeeds in lowering the cost of energy exchange and optimum energy distribution, thereby reducing the overall system cost. In addition, every MG has an energy storage system (ESS), and an analysis has been conducted to determine whether it affects line losses as well as the costs associated with buying and selling energy for each MG. The proposed HCM and the conventional non-coordination model (NCM) are compared. Effects are assessed, and a comparison indicates how successful the suggested HCM is. Results suggest that, in comparison to NCM, HCM has decreased electrical line losses. The results demonstrate that, in comparison to NCM, the proposed HCM model is a more economical means of facilitating energy exchange between microgrids; consequently, the model would significantly improve system efficiency. Power line losses were reduced by 69.7% because of the ESS integration, compared to conventional NCM. As a result of the enhanced energy exchange between MGs, HCM caused the energy purchasing costs to drop by 17.7%. Additionally, the pricing mechanism has been performing well and has improved the power exchange between MGs, which has helped to lower the load on the MG.

Electricity theft remains a significant challenge for modern power systems causing major financial damage to utility providers. With increasing adoption of advanced metering infrastructure in smart grids, energy efficiency and real-time monitoring have improved but the vulnerability of smart meters to tampering and cyber manipulation introduces new threats. To address these concerns at scale, this study proposes a cloud-based machine learning framework for intelligent electricity theft detection in residential sectors. The cloud-based setup, utilizing a Google Colab environment, facilitates centralized data processing, streamlined deployment, and real-time decision support. This study systematically resolves several critical limitations in existing approaches. Firstly, it addresses the severe class imbalance in real-world datasets, which significantly underrepresent theft cases compared to honest consumption. The synthetic minority oversampling technique is used to equalize representation and enable fair learning. Second, recognizing that synthetic oversampling can introduce noise and class overlap, a post-balancing noise reduction mechanism is integrated to refine the dataset and eliminate ambiguous samples. Third, by combining the interpretability of Decision Trees (DT) and the robustness of Random Forests (RF), an ensemble model named DTRF is proposed to enhance variance control and generalization. To improve the predictive performance of DTRF, the Firefly Algorithm (FA) is used to smartly adjust the model’s settings based on the specific features of the dataset. The proposed models are tested on two publicly available real-world datasets, Pakistan Residential Electricity Consumption (PRECON) and State Grid Corporation of China (SGCC), showing consistent improvements over standard models in various measures. On the PRECON dataset, the DTRF model outperforms baseline models and achieves a 25.68% reduction in log loss, a 31.23% decrease in Hamming loss, a 3.52% improvement in Cohen’s Kappa (CK), and a 4.10% increase in Jaccard score. The proposed DTRF-FA model further achieves a 40.97% reduction in log loss, a 36.38% decrease in Hamming loss, a 2.71% improvement in CK, and a 3.17% increase in the Jaccard score relative to the DTRF model. On the SGCC dataset, the DTRF model outperforms baselines with a 33.34% drop in Hamming loss, a 3.86% increase in CK, a 3.70% rise in the Jaccard score, and a 1.92% gain in the F1-score. The DTRF-FA model achieves a 75.00% reduction in Hamming loss, a 5.63% gain in CK, a 5.82% rise in Jaccard score, a 2.95% improvement in F1-score, and a 5.39% increase in Matthews correlation coefficient over the DTRF baseline. These improvements affirm the models’ ability to balance sensitivity and specificity while maintaining low false positive and false negative rates. Finally, to promote interpretability and transparency essential for stakeholder trust and regulatory compliance, the framework integrates two powerful explainable artificial intelligence techniques. Local Interpretable Model-agnostic Explanations (LIME) is used to generate local, instance-level explanations of individual predictions, while SHapley Additive exPlanations (SHAP) provides a global view of feature importance using cooperative game theory. By combining LIME and SHAP, the framework helps practitioners understand models’ decisions in detail and as a whole ensuring that it is accurate, clear, adaptable, and suitable for use in the smart grid settings.

One of the most complex and prevalent diseases is heart disease (HD). It is among the main causes of death around the globe. With changes in lifestyles and the environment, its prevalence is rising rapidly. The prediction of the disease in its early stages is crucial, as delays in diagnosis can cause serious complications and even death. Machine learning (ML) can be effective in this regard. Many researchers have used different techniques for the efficient detection of the disease and to overcome the drawbacks of existing models. Several ensemble models have also been applied. We proposed a stacking ensemble model named NCDG, which uses Naive Bayes, Categorical Boosting, and Decision Tree as base learners, with Gradient Boosting serving as the meta-learner classifier. We performed preprocessing using a factorization method to convert string columns into integers. We employ the Synthetic Minority Oversampling TEchnique (SMOTE) and BorderLineSMOTE balancing techniques to address the issue of data class imbalance. Additionally, we implemented hard and soft voting using voting classifier and compared the results with the proposed stacking model. For the Artificial Intelligence-based eXplainability of our proposed NCDG model, we use the SHapley Additive exPlanations (SHAP) technique. The outcomes show that our suggested stacking model, NCDG, performs better than the benchmark existing techniques. The experimental results of our proposed stacking model achieved the highest accuracy, F1-Score, precision and recall of 0.91, 0.91, 0.91 and 0.91, respectively, and an execution time of 653 s. Moreover, we have also utilized K-Fold Cross-Validation method to validate our predicted results. It is worth mentioning that our prediction results and their validation strongly coincide with each other which proves our approach to be symmetric.

Intrusion detection in Internet of Things (IoT) environments is a critical yet challenging task due to the heterogeneous nature of devices and the complex attack landscape. Traditional machine and deep learning models often suffer from limitations such as poor class balance, irrelevant or redundant features, suboptimal classification accuracy, ineffective hyperparameter tuning, and scarcity of labeled data. To address these challenges, we propose an enhanced intrusion detection system model. Variance threshold is applied to select informative features, while the proximity weighted random affine shadow sampling technique is used to balance the dataset effectively. Capsule Network (CapsNet) is employed for robust classification due to its ability to capture spatial hierarchies in data. To further optimize CapsNet, we implement the Reptile Search Algorithm (RSA), resulting in the Reptile-Optimized Capsule Network (ROC-Net). ROC-Net is further enhanced using Margin-Based Active Learning (MBAL), forming Marginal Active-learning with Reptile-optimized Capsule Network (MARCO-Net), which efficiently annotates the most uncertain samples from the unlabeled pool. Experimental results show that the proposed ROC-Net and MARCO-Net significantly outperform traditional models, achieving improvements of 8.75 % and 12.5 % in accuracy, 13.75 % and 11.25 % in precision, 1.19 % and 9.52 % in recall, $\mathbf{7. 4 1 \%}$ and $\mathbf{1 1. 1 1 \%}$ in F1-score, $\mathbf{1 1. 9 4 \%}$ and $\mathbf{2 0. 9 0 \%}$ in Matthews Correlation Coefficient, and 13.64 % and 22.73 % in Cohen's Kappa. Additionally, there is a reduction of 36.84 % and 52.63 % in hamming loss, and 75.86 % and 82.70 % in log loss, respectively. These findings demonstrate the effectiveness of the proposed system for accurate and efficient intrusion detection in IoT environments.

Hypertension continues to be a primary cause of global death, necessitating early and accurate forecasting for effective treatments. The existing methods have drawbacks such as class imbalance, poor modeling of sequential and spatial connections, high computation costs, and lack of interpretability, even though Deep Learning (DL) models offer possible solutions. To tackle these difficulties, we present VahigoNet, a novel blending DL model that incorporates vanilla recurrent neural networks (VRNN) for capturing temporal correlations, Google network for extracting hierarchical spatial features, and highway networks (HighwayNet) for adaptive feature refinements. To achieve strong generalization, we utilize the synthetic minority oversampling technique (SMOTE) for data balance. VahigoNet substantially outperforms baseline models, showing enhancements of 9.39% in accuracy, 10.27% in precision, 8.63% in recall, 9.39% in F1-score, and 3.10% in area under the curve-receiver operating characteristic. A 10-fold cross validation method is utilized to assess the model's generalizability, markedly reducing overfitting and improving robustness. A paired t-test is performed to evaluate statistical significance, demonstrating that the enhancements are substantial and clinically relevant. Additionally, explainable artificial intelligence (AI) methodologies, including local Interpretable model-agnostic explanations (LIME) and SHapley Additive exPlanations, are incorporated to provide both local and global perspectives on feature contributions. These explainability strategies enhance transparency, making VahigoNet a more interpretable and clinically reliable model for hypertension prediction. The results demonstrate that VahigoNet is an exceptionally efficient and transparent method, achieving a balance between predictive capability and practical relevance in medical diagnostics.

The Internet of Urban Things (IoUTs) regularly generates large amounts of data, making it a focus of cyberthreats such as denial-of-service attacks and malware bot networks. Traditional intrusion detection systems struggle to detect intricate attack patterns, handle class imbalance, capture temporal dependencies, and exhibit transparency. To address these limitations, we introduce a novel deep machine learning model, DeepSecure, a hybrid model that combines Deep Belief Networks (DBN) for hierarchical feature extraction and Deep Neural Networks for attack classification. DBN is used for feature selection through unsupervised learning to extract hierarchical representations in the IoUTs network data. We assess the random oversampling examples technique to improve model generalization and prevent class imbalance. DeepSecure is implemented using the TON_IoT dataset, which includes multi-source attack data indicative of Industry 4.0 cyber risks. In multi-class classification, DeepSecure shows improvement score of 21.25% in accuracy, recall, F1-score, and 18.29% in precision. Whereas, in binary classification, it increases accuracy by 11.23%, precision by 15.11%, F1-score by 10%, and recall by 5.32%. ANOVA T-test and 10-fold cross-validation are utilized for results validation to ensure DeepSecure's reliability. Additionally, we use Shapley additive explanations to interpret the DeepSecure's decision-making process to provide insight into feature contributions and model transparency. By effectively tackling IoUTs-specific cybersecurity challenges such as attack pattern detection, data imbalance, and lack of interpretability, the results demonstrate that DeepSecure is a practical, and transparent, solution for IoUTs network security.

In this study, we introduce two innovative hybrid models designed for predicting individual pharmacy costs: the Autoencoder-Gated Recurrent Unit (Auto-GRU) and the GoogLeNet-Residual Network (GR-Net). Utilizing data from high utilizers obtained through the Medicaid rebate program, these models aim to provide accurate predictions of total pharmacy costs for individual patients. Our approach involves rigorous data preprocessing, including the removal of missing values, and the fine-tuning of hyperparameters using the Adam optimizer.We systematically evaluate and compare the performance of these hybrid models with that of four individual models, Autoencoder (AE), Gated Recurrent Unit (GRU), GoogLeNet, and Residual Network (ResNet), using performance metrics such as Mean Squared Error (MSE), Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE), training time, inference time and memory usage. Without correlation, Auto-GRU outperformed its individual models with an MSE value of 0.1675 versus GRU’s MSE of 0.2806 and AE’s MSE of 0.4086. GR-Net, without correlation, also beats the individual models with an MSE value of 0.0027 versus GoogLeNet’s MSE value of 0.0106 and ResNet’s MSE value of 0.0201. Similarly, with correlation, Auto-GRU outperformed its individual models with an MSE value of 0.2655 versus GRU’s MSE of 0.4089 and AE’s MSE of 0.4243. GR-Net, with correlation, also beats the individual models with an MSE value of 0.0097 versus GoogLeNet’s MSE value of 0.0882 and ResNet’s MSE value of 0.0106. Both the hybrid models outperformed the individual models in terms of MAE, MAE and RMSE both with and without correlation. Also, to interpret models prediction, we have implemented Local Interpretable Model-agnostic Explanations, an eXplainable Artificial Intelligence (XAI) technique. These findings highlight the robustness and effectiveness of the hybrid models in predicting pharmacy costs, underscoring their potential for integration into healthcare expense management systems.

Accurate predictive models are crucial for early detection and intervention of Heart Disease (HD), which continues to be a major cause of death worldwide. However, the challenges of high dimensionality and data imbalance affect the precision and generalizability of predictions. We propose a novel deep model, the Fully Connected Wave Network (FCW-Net), for early HD prediction. To mitigate class imbalance, Proximity Weighted Synthetic (ProWSyn) oversampling technique is employed, while Principal Component Analysis (PCA) is used to reduce dimensionality, enhancing model efficiency and prediction accuracy. With an AUC-ROC of 0.9622, an accuracy of 0.9237, a precision of 0.8856, an F1-score of 0.9268, and a recall of 0.9721, our results demonstrate that the FCW-Net deep model achieves superior performance than baseline models when using PCA and ProWSyn. Model transparency is further improved by eXplainable Artificial Intelligence technique, SHapley Additive exPlanations analysis, which provides insights into feature contributions to HD risk.

Background: Diabetes remains a major global health challenge, contributing significantly to premature mortality due to its potential progression to organ failure if not diagnosed early. Traditional diagnostic approaches are subject to human error, highlighting the need for modern computational techniques in clinical decision support systems. Although these systems have successfully integrated deep learning (DL) models, they still encounter several challenges, such as a lack of intricate pattern learning, imbalanced datasets, and poor interpretability of predictions. Methods: To address these issues, the temporal inception perceptron network (TIPNet), a novel DL model, is designed to accurately predict diabetes by capturing complex feature relationships and temporal dynamics. An adaptive synthetic oversampling strategy is utilized to reduce severe class imbalance in an extensive diabetes health indicators dataset consisting of 253,680 instances and 22 features, providing a diverse and representative sample for model evaluation. The model's performance and generalizability are assessed using a 10-fold cross-validation technique. To enhance interpretability, explainable artificial intelligence techniques are integrated, including local interpretable model-agnostic explanations and Shapley additive explanations, providing insights into the model's decision-making process. Results: Experimental results demonstrate that TIPNet achieves improvement scores of 3.53% in accuracy, 3.49% in F1-score, 1.14% in recall, and 5.95% in the area under the receiver operating characteristic curve. Conclusions: These findings indicate that TIPNet is a promising tool for early diabetes prediction, offering accurate and interpretable results. The integration of advanced DL modeling with oversampling strategies and explainable AI techniques positions TIPNet as a valuable resource for clinical decision support, paving the way for its future application in healthcare settings.