Cardiovascular disease after post-COVID has become a life-threatening, deadly disease. A major percentage of the mortality rate occurring every year is due to heart-related diseases. India, being a middle-income nation, is facing a severe need for awareness and resources to reduce the untimely death, especially in the middle-aged population, due to cardiac arrest. Machine learning has been acting as an essential tool to predict an early occurrence of this fatal disease. There is a need to provide personalized recommendations to the patient if the patient is predicted to have heart disease. The paper aims at providing the various content-based recommendations to the patients based on 5 parameters: age, FBS, trestbps, chol, thalach, and CP so that the precautionary measures need to be taken by the person based on the recommendation provided by the model. The contribution of this work is summarized into three parts. a) A stacking ensemble-based meta-learner is developed to predict a person with heart disease. b) A machine learning pipeline is incorporated to automate the workflow of disease detection and c) A novel personalized recommendation method with respect to cardiovascular risk reduction and preventative measures providing the optimal solutions to the person suffering from heart disease and its prognosis. The proposed method is validated by providing the necessary recommendations to the patients, demonstrating significant risk reduction for individuals with high CVD risk. The result evaluation is done using a t-test showing a statistically significant level and an ROC curve with a value of 0.98 and a sensitivity analysis with a value of 0.91. The mean average precision (MAP) value is 0.75. Creating a machine learning-based model that forecasts the likelihood of heart disease onset is the aim of this research. The existence or absence of heart illness, given as a binary classification (0 = no heart disease, 1 = heart disease), is the outcome variable for this prediction task.

Mobile agents are powerful advanced technology for distributing computing with intelligent and autonomous execution of assigned tasks. During the execution, a mobile agent utilizes a specific life cycle in a malicious environment. Collaboration of mobile agents for securing medical data like patient information, patient medical report etc. plays an important role. In the digital world, smart health care system is required. For secure accessing of patient information from one hospital to another hospital a framework is proposed based on the variable threshold Chinese remainder theorem with the present and pride lightweight cipher. The Chinese remainder theorem is used for secret sharing of key among hospitals. The VT-CRT approach dynamically modifies the threshold in response to system variables, improving adaptability and security while minimizing computing cost. Present and pride are used for encryption and decryption of electronic medical records. This is the collaborative approach of mobile agents for creation and recreation of secret keys among hospitals for authentication. The effectiveness of the healthcare system is evaluated by the ability to securely and effectively transmit medical information. The lightweight PRIDE and PRESENT encryption methods provide efficient, low-latency encryption that is ideal for resource-constrained medical devices. The framework is compared to popular cryptographic systems like AES, DES, and Blowfish in terms of time complexity, communication cost, flexibility, and memory usage. Experimental results show that the proposed framework improves security while maintaining optimal speed, making it a reliable solution for safe patient information sharing in medical applications.

Predicting medical insurance costs is a difficult task that requires calculating future medical expenses for individuals or groups based on their personal and medical data. Deep learning is the robust technique that can extract complicated relationships and patterns from huge and varied data sources. In this article, we suggest a novel deep learning model to predict the cost of medical insurance for a specific person based on their age, BMI, sex, number of children and smoking status. First, the Z-score pre-processing technique is employed to remove the noisy data. Then the metaheuristic optimization algorithm modified Fire Hawk Optimization (BFHO) is introduced for feature selection (FS) to select the most relevant features data, thus decreasing the number of features. Additionally, the dynamic chunk-based max pooling (DCMP) technique is employed to improve the pooling layer in CNN network and the improved golden eagle optimization (IGEO) approach is utilized to enhance the weight of the CNN network. Finally, this improved CNN is used for the prediction of medical insurance based on the medical dataset. The predictive performance of the proposed approach is systematically evaluated and benchmarked against several traditional methods, including the Improved Whale Optimization Algorithm (IWOA), Fire Hawk Optimization (FHO), Improved Manta Ray Foraging Optimization (IMRFO), Honey Badger Algorithm (HBA), and Binary Grey Wolf Optimizer (BGWO). The experimental results show that the proposed model is the best approach for the cost prediction of medical insurance.

Early identification of osteopenia, an easy indicator of osteoporosis, is essential for fracture preventative measures and the restoration of long-term bone wellness. This research sets up and assesses a machine learning screening framework designed for detecting the beginning of osteopenia, utilizing enhanced clinical datasets that include demographic, medical, and bone health variables. The Custom Weighted Ensemble Model, through a comparative analysis with different classification algorithms, obtained the highest accuracy of 98.69%, surpassing standard models including Random Forest and Gradient Boosting, both at 98.48%. The ensemble approach demonstrates enhanced precision and F1-scores, underscoring its effectiveness in balancing both precision and sensitivity compared with prior research which concentrates primarily on osteoporosis. This investigation demonstrates the initial stage of osteopenia detection along with it suggests that optimized ensemble strategies significantly improve predictive performance. The findings indicate the potential of data-driven models to enhance preventive bone health care and decrease the risk of future fractured bones. Early recognition of osteopenia—widely noticed as the impassive initial level for osteoporosis—is essential for safeguarding the health of bones and avoiding severe fractures. This paper presents a detailed review of existing methodologies for diagnosing and predicting osteopenia, with a focus on the comprehensive overview of machine learning (ML). By combining insights from significant studies on osteoporosis prediction and choosing a comprehensive dataset incorporating with demographic, medical, and bone health parameters, we reveal essential predictive factors. Additionally, we offer a state-of-the-art framework for creating reliable machine learning models specifically suited for early detection. This study highlights how data-driven innovation can improve clinical care, slow the progression of osteopenia, and pave the way for a future free of fractures.

Social work practice faces critical challenges in accurately identifying high-risk families for timely intervention due to limited labeled data. This research develops an intelligent system integrating supervised and semi-supervised learning for early warning, coupled with uncertainty-aware resource optimization. We propose a hybrid framework combining Self-training and Co-training leveraging 1,200 labeled and 13,800 unlabeled family cases. The proposed method achieves 85.6% accuracy and 82.4% recall on the high-risk class, with 95% confidence intervals, substantially outperforming supervised-only baselines. A multi-objective allocation model balances effectiveness, cost, and equity while incorporating prediction uncertainty via robust optimization. Sensitivity analysis over 500 bootstrap scenarios demonstrates service quality guarantees. Subgroup fairness audits show consistent recall across demographic groups (max gap 1.2%). The optimized allocation reduces manpower by 22% while improving intervention success from 68.5% to 84.2%. SHAP-based explainability and human oversight support ethical deployment in social services.

Student dropout is still a crucial issue in many colleges and universities in countries including Indonesia despite various initiatives that have been implemented to reduce its rate. As a preventive effort, many educational institutions continuously monitor several factors that potentially affect student graduation and dropout rates. However, the increasing ratio of student numbers to administrative staff has made manual processes inefficient. This paper presents a novel method to monitor student learning outcomes using a subgraph matching approach to identify students who potentially drop out early or fail to graduate due to low learning achievement. This study emphasizes diagnostic accuracy in identifying students at risk of dropping out as the main focus of the proposed contribution. Performance assessment is directed at the model's ability to accurately distinguish between at-risk and not-at-risk students, particularly through evaluations that emphasize the minority class. This approach strengthens the model's role as the foundation of a more reliable early warning system relevant to supporting academic decision-making. In this research, the academic achievement of each student is represented as a graph that represents the relationship between each student and courses that have been enrolled in the previous semesters. By comparing the learning achievement pattern between each student and those students who have either "graduated" or "dropped out", the students with elevated dropout risk can be identified in a computationally efficient manner.  The experiment findings showed that the proposed method can identify students with elevated dropout risk with 89.47% average accuracy. Graph-based approaches are capable of capturing structural relationships and complex interaction patterns among student activities that cannot be explicitly represented by other models. Although subgraph matching is known to have high computational complexity, the identification process can be accelerated through subgraph pattern constraints, preprocessing strategies, and search optimizations, thereby supporting claims of efficiency both theoretically and practically.

Epilepsy is a prevalent neurological disorder characterized by recurrent epileptic seizures. After a stroke, it is a highly common neurological condition. For identifying seizures, the Electroencephalogram (EEG) is the gold-standard modality for capturing and recording the brain's electrical activity with high temporal resolution. The existing machine learning approaches have been leveraged before the emergence of deep learning. However, these models limited the performance as they included handcrafted features. The deep learning models perform the feature extraction automatically, which improves the classification accuracy compared with the conventional strategies. Therefore, in this article, an epileptic seizure detection approach on the basis of deep learning methods is developed to identify the abnormal brain functionalities in the initial stage. Initially, the EEG signal collection is carried out through standard benchmark databases. After that, the spectral and statistical features are directly extracted from the acquired EEG signals. Subsequently, the Short-Time Fourier Transform (STFT) is applied to transform the EEG signals into spectrograms, and then the features from those spectrograms are extracted via Vision Transformer (ViT). Furthermore, the Wave-based features are also extracted, which provide a comprehensive view of the brain activities. After that, a Coordinate Attention-based Feature Fusion mechanism is employed to fuse these diverse feature types, which captures complementary details from the EEG signals. The fused features are subjected to the Adaptive Dilated dense Recurrent neural Network with a Novel Activation Function (ADRNet-NAF) for the Epileptic seizure detection. The detection accuracy of the proposed model is improved by optimizing the hyperparameters of the ADRNet-NAF model using Modernized Exploration of Magnificent Frigatebird Optimization (MEMFO) during network training. The empirical analysis over the traditional methods is conducted to validate the Epileptic seizure detection performance of the proposed model using various measures. The proposed model achieves 94.98% accuracy, 89.30% sensitivity and 99.08% specificity.

We present GreenCloud-RL, a formally defined scheduler for multi-cloud fleets that jointly optimizes carbon emissions, operational cost, and service-level agreement (SLA) risk. The scheduling problem is modeled as a constrained Markov decision process (CMDP) in which expected cost–carbon reward is maximized subject to aggregate SLA-risk and power-cap constraints. The framework integrates (i) a physics-informed DVFS power model, (ii) queueing-theoretic latency predictors, (iii) probabilistic short-term forecasts of regional grid carbon intensity and electricity price with uncertainty penalties, and (iv) a dual-variable actor–critic algorithm that provably enforces constraint satisfaction online. We analyze the algorithm’s convergence under bounded rewards and provide distributed complexity bounds for large-scale multi-cloud control. Empirical evaluation on production-like workloads shows that GreenCloud-RL reduces SLA-violation rate by 37.6% and power-cap violations by 53.8%, while lowering mean tardiness to 1.9 s and improving throughput by 4.2% compared with strong baselines. A region×regime analysis (load ∈ {low, medium, high}; grid ∈ {clean, neutral, dirty}) demonstrates consistent reductions in both cost and carbon emissions without degrading SLA satisfaction. Post-hoc calibration improves reliability (ECE 4.8%→1.7%, lower Brier score), enabling trustworthy on-time probability estimates for risk-aware admission and tenant-tier guarantees. Scalability tests show a p95 scheduling latency of ≈41 ms at 100k jobs, ≈0.25 ms inference per decision, modest CPU/GPU overhead, and ≈5 s controller failover. These results establish, both theoretically and empirically, that multi-objective optimization of carbon, cost, and reliability is achievable in production-scale cloud environments while maintaining contractual SLA guarantees. 

Predicting Carbon dioxide (CO₂) emissions is essential for guiding environmental policies and climate strategies, yet accurate local-scale prediction remains challenging due to complex spatial and temporal factors. Existing approaches often focus on regional or global models and do not fully exploit high-resolution satellite imagery for local trend analysis. To overcome these limitations, Conditional Pine Cone Optimization enabled Attention-Based Long Short-Term Memory (CPCO_ALSTM) is devised for CO2 Emission prediction for Trend Analysis in Satellite Image. In this approach, satellite images of the query region acquired at different time intervals are used as input. Initially, a Kalman Filter is employed for noise reduction and temporal smoothing during preprocessing. Then, segmentation of satellite image is done using Dense-Res Recurrent Prototypical Network (DRRP-Net), wherein DRRP-Net is developed by combining Dense-Res-Inception Network (DRINet) and Recurrent Prototypical Network (RP-Net). Based on the segmented regions and extracted features, carbon emission estimation is carried out. The estimated CO₂ emissions are then validated against ground-truth emission data obtained from a reference database to assess prediction accuracy. For temporal trend analysis, an Attention-Based Long Short-Term Memory (ALSTM) model is utilized to capture long-range dependencies in emission patterns. The ALSTM parameters are optimally tuned using the proposed CPCO algorithm, which combines Pine Cone Optimization (PCO) with Conditional Autoregressive Value at Risk (CAViaR) to enhance convergence stability and prediction robustness. Additionally, CPCO_ALSTM has attained 96.35% of accuracy, 0.087 of normalized MSE, and 0.957 of Jaccard Coefficient.

Adequate nutrition, patient acceptance and consistent dietary adherence are required for optimal chronic glycaemic control in T2DM. Customised and clinically adapted meal planning may best support these factors. This work is situated in the field of Medical Informatics, combining explainable sequence modelling with clinically meaningful dietary restrictions, ensuring both prediction accuracy and health-oriented decision support for Type 2 Diabetes Mellitus (T2DM). In contrast to the traditional RNN based recommenders, the proposed approach explicitly incorporates adherence-aware assessment, interpretability and nutritional safety. In this paper, we propose CA-Seq2Seq-LSTM-Attn, a Constraint-Aware Seq2Seq-LSTM with Dot-Product Attention framework for generating customised seven-day meal plans (21 meals) from nutritional constraints and previous dietary behaviour. We use an LSTM-based Seq2Seq architecture in place of Transformer-based models because of its strong inductive bias for modelling short, structured temporal sequences (21 meals over 7 days), better generalisation on small high-variance dietary datasets, and lower computational complexity than data-hungry Transformer models. The model is a sequence-to-sequence Long Short-Term Memory (LSTM) architecture with attention for capturing temporal eating patterns in daily meal sequences. The constraint-aware decoding strategy ensures that the calorie, carbohydrate, sugar, protein limits, and the Glycaemic Index (GI) and Glycaemic Load (GL) thresholds are satisfied, which is crucial in the context of diabetes management. The attention mechanism makes the respective forecast more interpretable by pointing to important previous meals. We evaluate the system using enriched user-recipe interactions and comprehensive nutritional data from the Food.com dataset. The experimental results show a good ranking performance with NDCG@10 of 0.5026 and Recall@10 of 0.5909. The model also ensures a better diet consistency, as indicated by the lower variance in carbs (826.6690), weekday adherence (74.39%), intra-plan variety (0.7180), and high explainability (minimum perturbation-based faithfulness confidence reduction: −0.0288). Incorporating temporal sequence modelling, attention-based interpretability and constraint-aware decoding substantially improve clinical nutritional feasibility and recommendation accuracy. The proposed framework outperforms the traditional Collaborative Filtering (BPR-MF, NCF) and sequential deep learning baselines (GRU4Rec, SASRec) on all ranking and nutrition metrics. The proposed CA-Seq2Seq-LSTM-Attn framework is a promising decision-support tool for personalised dietary control in T2DM as it generates interpretable, nutritionally balanced and clinically matched meal recommendations.

Clinical brain MRI scans for Alzheimer's disease diagnosis often suffer from motion artifacts, signal dropout, and incomplete acquisitions. While deep learning methods like LaMa produce visually coherent inpainting, they may alter critical anatomical structures, and traditional methods such as OpenCV Telea maintain local continuity but introduce excessive smoothing. This study presents a dual-stage generative adversarial network with attention gates for medical image restoration, uniquely complemented by a novel gradient-weighted hybrid blending strategy that adaptively combines GAN outputs with classical inpainting based on local image gradients. The architecture employs an attention-enhanced U-Net generator and refinement network, supervised by patch and global discriminators through combined adversarial, reconstruction, perceptual, structural similarity, and gradient losses. Evaluation on 12,491 balanced Alzheimer's MRI scans across four severity stages with synthetic 10–30% occlusions shows that the proposed GAN reduces mean squared error by 48.7% versus LaMa and 41.3% versus OpenCV, achieving 17.44 dB PSNR and 0.9796 SSIM (computed over the brain region). The gradient-guided Hybrid-GAN maintains reconstruction quality with a total inference time of approximately 8.97 ms per image. Downstream VGG16 classification reveals Hybrid-GAN preserves diagnostic information most effectively, attaining 96.05% accuracy, only 0.96 points below the 97.01% baseline and surpassing all alternative methods. These results demonstrate that attention-driven generative models combined with structure-aware blending provide a practical and novel solution for artifact mitigation in neuroimaging diagnostics.

Stock movement prediction is a challenging task due to the complex temporal dynamics of financial time series and the interdependence among stocks. Most existing approaches either model each stock independently or rely on a single static graph, which limits their ability to capture heterogeneous and time-varying relationships in financial markets. To address this issue, this paper proposes a dynamic dual-graph GCN-GRU framework for next-day stock movement prediction. The proposed model integrates two complementary graph structures: a static industry graph to encode long-term sectoral relationships and a dynamic correlation graph to capture evolving co-movement patterns among stocks over time, where the dynamic graph is constructed in a causally consistent manner using only historical information available before the target prediction. For each graph, a graph convolutional network (GCN) is employed to learn relational stock representations, while a gated recurrent unit (GRU) is used to model temporal dependencies from sequential graph-based embeddings. The two graph-specific representations are integrated through an adaptive gated fusion mechanism. By jointly exploiting structural market information and temporal dynamics, the proposed framework provides a unified representation for next-day stock direction forecasting. Experiments are conducted on two public stock datasets constructed from the S&P 500 and VN30 markets using engineered features derived from daily trading data. The evaluation considers both standard classification metrics and finance-oriented indicators, including IC, Sharpe ratio, and maximum drawdown. The empirical results demonstrate the effectiveness of the proposed approach compared with conventional machine learning, sequence-based, and graph-based baselines. These findings suggest that combining dynamic relational modeling with temporal learning is a practical and reproducible forecasting framework for stock movement prediction in both developed and emerging markets.