About
“My name is Md Mehedi Hassan Melon. I hold a Bachelor’s in Automation Engineering from Nanjing Tech University and an MBA from the International American University, USA. Currently, I am pursuing a Ph.D. in Information Technology at the University of the Cumberlands, specializing in Artificial Intelligence. My research interests include AI, data science, machine learning, IT in business, and automation.”
Experience
No experience added yet.
Education
University Of the Cumberlands
PH.D
International American University
MBA
Nanjing Tech University
Bachelor
Skills
Other Publications
AI-Driven Trust and Bio-Inspired Optimization Integration in Standardized Digital Twin Frameworks for Next-Generation
IEEE Communications Standards Magazine, May 2026
The development of next-generation communication systems, including 6Gs and beyond, requires smart, reliable, and standardized architectures to manage network complexity. The paper provides a foundation for a hybrid AI-based trust and bio-inspired optimization method in a standardized Digital Twin (DT) framework of secure and adaptive networks and safe communications. The system proposed follows the principles of AI-based trust modelling and enhances node reliability and uses bio-inspired optimization algorithms, namely Particle Swarm Optimization (PSO), and Ant Colony Optimization (ACO), to efficiently allocate resources and adapt the network. Relative evaluations of a prototype Digital Twin architecture relative to a baseline standardized DT architecture without embedded trust or optimization units advocate improvement of 13.7% trust accuracy, 22.8% latency risk-reduction, 18.5% energy expenditure and 36% interoperability across a heterogeneous 6G network with dynamic topology. The framework complies with new IETF and Digital Twin Consortium (DTC) standards and enables scalable, interoperable communication environments across heterogeneous networks. These findings provide the basis on which secure, self-optimizing, and standardized DT ecosystems can be deployed in future communication networks.
Risk-informed flood management under climate variability: A hybrid temporal–ensemble artificial intelligence framework
Bulletin of Atmospheric Science and Technology, Apr 2026
One of the most significant meteorological risks to South Asia’s infrastructure, livelihoods, and prosperity is flooding. The lack of ground truth and the significant temporal variability of flood risk in data-scarce regions make credible flood risk prediction difficult, as does the need for an interpretable decision-support system. This research presents a mixed artificial intelligence system for monthly flood risk prediction under climatic fluctuations, prioritizing decision relevance over predictive accuracy. A Tree-Based Interpretable Ensemble of Flood Risk Management (IBE-FRM) with XGBoost, LightGBM, and Random Forests, and a Hybrid Temporal Ensemble Decision Support Model (HTEDSM) with LSTM-based sequence learning were created from long-term monthly rainfall and temperature records of Rajshahi, Bangladesh. Applying data-efficient synthetic labeling with percentile thresholds of cumulative rainfall permitted supervised learning and was not confined to whole flood records to estimate flood-prone months. Cumulative rainfall is the most critical driver of monthly flood danger, and lagged and rolling rainfall features provide the best forecasts. The conservative and dependable IBE-FRM model delivers unambiguous and policy-relevant flood risk estimates for daily monitoring and stakeholder communication. Conversely, HTEDSM is more sensitive in time; it records delayed climatic influences that are significant for early warning and preparedness for emergent flood dangers. The discrimination metrics and temporal risk change comparison show that the models have an AUC-ROC of 0.68 and 0.81, respectively, indicating the complexity of climate-driven flood forecasting and the effectiveness of time-based modeling. Overall, the suggested hybrid model demonstrates that combining interpretable ensemble learning with temporal deep learning can provide a balance of robustness, explainability, and the ability to offer early warning. The conclusions suggest that hybrid artificial intelligence systems can enable risk-informed flood control decisions in climatic uncertainty, especially in data-scarce contexts.
Attention-guided fusion of swin transformer and DINOv2 features for peripheral blood cell classification across eight classes
Discover Artificial Intelligence, Apr 2026
Accurate classification of peripheral blood cell types from microscopy images can support automated hematology workflows by reducing manual review burden and improving consistency. However, fine-grained recognition remains challenging due to strong inter-class similarity, intra-class morphological variability, and confounding cytoplasmic texture patterns that blur decision boundaries among related cell types. We propose SwinDiNO-AFNet for eight-class classification of basophil, eosinophil, erythroblast, immature granulocyte (IG), lymphocyte, monocyte, neutrophil, and platelet. SwinDiNO-AFNet is a dual-path hybrid that fuses hierarchical window-based features from a Swin Transformer with global representations from a pretrained DINOv2 vision transformer used strictly as a feature extractor. The flattened Swin and pooled DINOv2 features are concatenated and refined through a channel attention module to obtain attention-weighted fusion before the final classifier. To ensure rigorous evaluation, the training pipeline enforces leakage control via perceptual-hash based near-duplicate removal before fold creation, with consistent RGB enforcement, 224 × 224 resizing, ImageNet normalization, and train-only augmentation and robustness noise applied exclusively within training folds. Under leakage-safe stratified fivefold cross-validation, SwinDiNO-AFNet achieves 98.61 ± 0.21% validation accuracy with 98.46% macro-F1, and statistical testing (Welch with Holm adjustment) supports improvements over multiple baseline backbones under the same protocol. Precision-recall analysis further indicates near-ceiling separability (micro-AUPRC 0.9979), while confusion matrix and Grad-CAM based error analysis show that remaining failures concentrate in morphologically ambiguous cases, including IG confusions toward granulocytic patterns. Overall, the results show that attention-guided fusion of complementary Swin and DINOv2 representations, combined with leakage-safe cross-validation and XAI-backed diagnostics, provides a strong solution for eight-class peripheral blood cell classification on the studied dataset.
Hybrid Deep Learning Framework for Multiclass Diabetic Retinopathy Classification from Fundus Images
IEEE, Mar 2026
Accurate grading of diabetic retinopathy from retinal fundus images is crucial for early intervention and prevention of vision loss, but automated systems have trouble with fine interclass variation and good generalization across disease stages. Existing deep learning approaches commonly rely on either convolutional or transtransformer-based former based representations alone, which limits their ability to jointly capture fine-grained local lesions and global contextual patterns. To overcome the above drawback, a hybrid classification scheme combining convolutional feature extraction with transformer based global representation learning is proposed, aided with a structured data augmentation pipeline to enhance the contrast while suppressing the noise. The model is evaluated on a publicly available multi-class diabetic retinopathy dataset consisting of five severity levels and evaluated with the help of confusion matrices, classification, accuracy curves, loss curves, and receiver operating characteristic analysis. The proposed framework obtains an overall accuracy of 99.83 percent with always high precision, recall, and F1 score for all classes and an area under the curve of 1.00, outperforming strong baselines such as VGG19 ConvNeXt V2 and InternViT. Robustness is demonstrated by stable patterns of convergence behavior minimal gap between generalization and clinically consistent patterns of confusion concentrated between neighboring levels of severity rather than distant classes. Although the proposed framework is a strong performer, the evaluation has only been done on one public dataset and additional multi-institutional validation of this technique is needed.
Self-Adaptive Artificial Intelligence Systems for Large-Scale Data-Driven Decision Making
IEEE, Mar 2026
The growing magnitude, diversity, and speed of data produced across contemporary cyber-physical, enterprise, and cloud-edge systems has revealed inherent constraints with the inflexible artificial intelligence (AI) models regarding providing real-time choices. Self-adaptive artificial intelligence systems have become an attractive paradigm to overcome these challenges through enabling continuous learning, dynamic model reconfiguration and context-specific optimization in the changing data distributions and operational constraints. In this paper, a detailed self-adaptive AI architecture to serve large-scale decision making with data is provided, which involves online learning, feedback-based adjustment, and automatic policy optimization. The suggested architecture takes advantage of the hybrid intelligence methods merging deep learning, reinforcement learning, and meta-learning to reach strong performance in non-stationary and uncertain environments. Adaptive control loops are incorporated to check data drift, system performance and the consumption of resources and recalibration of a model can be done in real time without involving human intervention. The comprehensive test on large-scale datasets indicates that the proposed solution is much more accurate in making decisions, flexible and resilient to systems against the traditional non-adaptable and semi-adaptive AI models. The outcomes confirm the effectiveness of self-adaptive intelligence in aiding scalable, dependable, and explainable decision-making of dynamic data-intensive applications including smart infrastructure, autonomous systems, and intelligent business analytics.
The Role of AI in Transforming Organizational Performance: A Review of Applications in Business Management
IEEE, Mar 2026
This review paper systematically examines the transformative role of artificial intelligence in enhancing organizational performance across business management functions. The objective is to synthesize recent literature on AI applications, their impacts on key performance metrics, associated challenges, and strategic implications. Employing a systematic literature review methodology—drawing from Scopus-indexed articles and bibliometric analyses published since 2020—we analyzed over 1,300 studies to uncover trends, themes, and gaps in AI-driven management practices [1]–[3]. Key findings highlight AI’s profound effects: machine learning, natural language processing, predictive analytics, and robotic process automation boost operational efficiency by automating routine tasks and reducing costs [4]–[6]; enhance decision-making accuracy through real time data insights and forecasting [7]–[9]; drive innovation and creativity in high-tech enterprises via synergistic human AI collaboration [10], [11]; and strengthen competitiveness by optimizing supply chains, personalizing marketing, and improving customer satisfaction [1], [12], [13]. The global AI market’s projected growth to $2,025 billion by 2030 underscores its strategic centrality [1]. However, barriers such as data privacy, ethical biases, skills gaps, and legacy system integration hinder adoption [3], [14]–[16]. This paper bridges fragmented research by offering a multilevel framework integrating AI’s holistic impacts, post-adoption dynamics, and multidisciplinary insights [10], [17]–[19]. Contributions include research agendas on human-AI interactions and policy recommendations for practitioners to foster ethical AI literacy, reskilling, and sustainable integration, enabling long-term performance gains [2], [20].
Explainable Ensemble Machine Learning Framework for Employee Attrition Forecasting: An Integrated SHAP-Based Approach to HR Decision Intelligence
IEEE, Jan 2026
The modern organizations are grappling with some very crucial challenges in anticipating the attrition of employees because they depend on the intuitively-based assessment method and survey-dependent methodology that cannot be predicted and interpreted. The paper presents a framework of explainable ensemble machine learning, which combines XGBoost and SHAP (SHapley Additive exPlanations) analysis to predict workforce turnover with the help of historical human resource, productivity, and engagement indicators. The methodology will include a comprehensive data preprocessing, feature engineering, construction of the ensemble model, and validation based on interpretability, which will be conducted in the form of ROC-AUC analysis, confusion matrices, and decomposition of SHAP values. Results of an experiment prove the 83 percent classification accuracy and competitive AUC performance (0.67) and the SHAP analysis shows that the EnvironmentSatisfaction, YearsWithCurrManager, and JobSatisfaction are the key predictors of the attrition. The framework has transparency-accuracy synergy, where human resource practitioners use evidence-based retention strategies to address high-risk groups of employees. It is an approach that helps to develop organizational decision-making, reduce recruitment costs, and improve the strategic workforce planning with the help of quantitative behavioral analytics in dynamic enterprise settings by converting opaque algorithmic-based predictions into workforce intelligence that can be acted upon.
ML based myocardial infarction detection from ECG
International Journal of Science and Research Archive, Dec 2025
Fast and reliable screening of the electrocardiogram helps clinicians catch cardiac problems early. We present a classical machine learning pipeline that separates myocardial infarction from confirmed normal 12-lead ECG recordings. The pipeline builds a compact set of statistical, spectral, wavelet, and heart rate variability features across the twelve leads, then trains a Support Vector Machine and a Random Forest on the PTB-XL corpus. On the held out fold we reach 0.944 accuracy with the SVM and 0.929 accuracy with the Random Forest, with AUROC values of 0.958 and 0.946. The ensemble also exposes which features drive the decision, which supports clinical interpretability. The approach stays light on compute and fits low resource settings where deep networks are hard to deploy.
Efficient graph-based intrusion detection for can bus security with minimal training data
IEEE, Dec 2024
In this study, we explore the efficacy of using a graph-based intrusion detection system (GBIDS) for Controller Area Network (CAN) bus security with minimal training data. Traditional machine learning models often require extensive data for training, typically around 70%, to achieve high accuracy and reliability. However, our approach demonstrates that even with significantly reduced training data—specifically, using just 30% of the available 70% of data (21% of total data)—we can maintain high levels of accuracy, precision, recall, and F1 score. This minimal data training is facilitated by on-the-fly training capabilities akin to edge computing, allowing for real-time adaptation and learning. Our results indicate that it is not always necessary to rely on the conventional 70% training data benchmark to ensure robust CAN bus security. By leveraging advanced graph-based algorithms and edge computing techniques, we achieve efficient and effective intrusion detection with reduced data requirements, thus offering a promising solution for real-time CAN bus security in resource-constrained environments.
Collaborator Notes
No notes yet.
Sign in to write a note.