Professor Celestine Iwendi

HALOGrid is an adaptive edge–cloud malware detection framework for IoT traffic. The approach couples a lightweight LSTM (residual paths, attention, drift-penalty regularization) for low-latency edge inference with a telemetry-driven tuner that performs real-time hyperparameter updates. The tuner employs Augmented Grid Search (AGS): a stage-wise coarse-to-fine exploration with stochastic perturbations, early-stopping of inferior candidates, validation-weighted corrections, and expectation-weighted deployment. A resynchronization controller blends edge and cloud states using divergence- and delay-aware gating; updates are secured via mTLS transport and signed artifacts with rollback. The pipeline integrates preprocessing, drift estimation over multi-metric streams, adaptive learning-rate/regularization adjustment, and A/B deployment safety. Evaluation on CICIoT2023 reports 98.74% accuracy, 1.21% false positive rate, and 12.8,ms mean inference latency on Jetson Nano; energy consumption averages 52.5,mJ/inference. Compared with SGM, HPAI, DFN, ODMS, MIHT, AIMO, IEMS, and DOFD, HALOGrid maintains higher detection fidelity with lower tuning overhead through AGS and secure edge–cloud refinement.

Humans face various diseases that are mainly caused by environmental conditions and living habits. These diseases exhibit several symptoms and can share a relationship based on their symptoms. The identification and interpretation of these groups of symptom-based diseases can aid in developing treatment plans for a new outbreak of disease. This research explores the intersection of machine learning and healthcare, specifically focusing on the enhancement of disease classification through symptom-based cluster analysis. By leveraging unsupervised machine learning algorithms, patterns and relationships within diverse symptom datasets were identified, revealing novel associations and subtypes in disease manifestation. The integration of a Large Language Model (LLM), specifically OpenAI's Generative Pretrained Transformer(GPT), played a pivotal role in interpreting and communicating the complex outputs of the machine learning process. The results indicated a significant improvement in defining distinct clusters based on the relationship between diseases and symptoms, with GPT-4o providing simplified explanations that bridge the gap between machine-generated insights and healthcare professional's understanding. The study's findings offer a more profound understanding of the distinctive features characterising the different clusters of diseases generated by the machine learning models. The healthcare field produces extensive and varied data, which machine learning algorithms can leverage to detect new illnesses and optimize treatment plans 1. Deep learning (DL), when trained on high-quality data, has significantly advanced clinical diagnostics and facilitated disease clustering 2. One example is symptom-based clustering, which can enhance diagnostic accuracy and support personalized patient care 3. Diseases with overlapping symptoms pose significant challenges for accurate clinical diagnosis, a problem that can be mitigated through coordinated care and collaboration between multidisciplinary teams 4. Traditionally, physical exams or laboratory tests are used to identify diseases. This process can be complicated and sometimes inaccurate, as many diseases share similar symptoms 5. ML-enabled techniques help to discover new disease subtypes and understand the diversity of the patient population by uncovering hidden patterns within complex data sets 6. Symptom-based cluster analysis is an effective technique for providing precise and targeted medical information 7. However, interpreting these complex models poses a unique challenge. Watson 8 argued that while clustering algorithms efficiently reveal connections, converting these clusters and patterns into meaningful medical insights is difficult.

Vehicular networks face increasing challenges to ensure security, privacy, and efficiency in dynamic communication environments. Current solutions often lack adaptability to evolving threats and efficient mechanisms for preserving privacy and reducing computational overhead. This study proposes PrivNet, a framework that integrates generative AI with advanced cryptographic and trust mechanisms to address these limitations. The framework comprises the Quantum-Augmented Holographic Cryptographic System (QAHCS) for dynamic and secure key generation, the Neural Overlap Privacy System (NOPS) for adaptive pseudonym morphing and entropy-driven identity obfuscation, and the Self-Supervised Generative Anomaly Detection (SS-GAI) module for real-time threat modeling and counter-anomaly injection. The system also incorporates Hyperledger Mesh for energy-efficient and secure transaction validation. Simulations were performed using NSL-KDD, CICIDS2017, and Car-Hacking / VeReMi datasets for 300 minutes. The results demonstrate a 12% improvement in detection accuracy, a 23% improvement in energy efficiency, and a 22% reduction in resource utilization.

Advancement in wireless and communication technologies has remarkably boosted healthcare services such as Medical WSN. Protecting the patient's health‐related data against malicious activities is also essential. It is mandatory to ensure its dependability and reliability. The reliability of the proposed model in Secure Multimedia Transmission for medical wireless sensor network (IRSMT) system considers the need for authentication and confidentiality in security. Additionally, it enhances the transmission reliability during multimedia data transmission compared to the prevailing methods. To provide reliable multi-media (MM) data transmission, an improved energy‐efficient protocol is considered where the protocol differentiates MM and non‐MM data to enhance routing methodology for MM transmission. The proposed IRSMT enhances adaptability by balancing media quality with prompt delivery and loss tolerance. It is achieved through the anonymous routing method, which maintains the node secrecy using the SHA 256 method. It reduces the probability of data retransmission and provides less processing delay to acquire routing reliability. The simulation results demonstrate the advantages of IRSMT in comparison with the prevailing protocols in performance metrics such as throughput, packet delivery ratio, jitter etc.

Waste detection and classification are critical processes in modern waste management systems, as they enable the efficient sorting and processing of various waste types. The significance of effective waste detection lies in its potential to minimize environmental impact and promote sustainability. With the growing volumes of urban waste, implementing advanced detection technologies is essential to facilitate timely interventions and optimize waste handling practices. For effective waste detection, several Deep Learning-based techniques have been implemented. Yet, they are limited by several drawbacks including poor accuracy, generalizability, computational efficiency, and class imbalance issues. This research work develops an automated methodology that incorporates Deep Convolutional Neural Networks with Adaptive Boosting for accurate waste detection and categorization. This study deployed a data augmentation step to enhance the quantity of images and resolve class imbalance problems. Deep Convolutional Neural Networks automatically identify and learn relevant features from raw image data, such as edges, textures, and shapes, which are crucial for distinguishing between different types of waste. The proposed model introduces an Adaptive Boosting ensemble learning technique for enhancing the classification performance by combining the outputs of several weak classifiers. Then, the proposed technique adopted a Binomial Crossover Ship rescue algorithm that incorporates the Ship Rescue Optimization algorithm with the Binomial Crossover Strategy that fine-tunes the hyperparameters of the proposed technique and improved the overall effectiveness of the model. In addition, the effectiveness of this study is evaluated using several waste detection datasets with distinct performance measures and the model attains superior detection results such as accuracy of 98.82% and precision of 98.56%. The simulation findings show that the proposed method provides an excellent contribution to early and accurate waste detection systems.

With the rapid changes in the mobile network environment and the dynamical user's interest, existing recommendation algorithms are unable to provide resources that meet user needs, which means both the accuracy and efficiency of resource recommendation are not good enough. Therefore, this article proposes a parallel recommendation algorithm for multi interactive resource based on label attributes and behavior sequences in mobile network. The proposed method first obtains users' preferences for resources based on label attributes to increase the accuracy of recommendation; and then computes the similarity between resources to remove the redundant resources and improve recommendation efficiency. Then, Deep Convolution Generative Networks (DCGN) is used to process interaction data between multiple users and resources. Here, the input interaction behavior sequence is fed into a dual Gated Linear Unit (GLU) , and Gated Recurrent Unit (GRU) based on attention mechanism is used to extract the change of user's interest. At the same time, a feature crossover module is used to learn the target resource connection to make recommendations more relevant. Finally, a Deep Convolutional Neural Network (DCNN) is used to output the user resource interaction score to complete the resource recommendation. Experimental results show that the Normalized Discounted Cumulative Gain (NDCG) and hit rate are 0.35 and 0.18 respectively when the length of recommendation list is 8, with minimum Logloss 0.2567 and maximum Area Under the ROC Curve (AUC) 0.9157, which means the coverage rate of proposed resource recommendation is high. The resource recommendation takes 46.72 seconds to process large-scale data, which indicates that the proposed algorithm has high recommendation efficiency.

This research study aims to develop and implement a sustainable and effective power solution for metropolises with a power demand ranging from 2.5 to 25 MW. The primary objective is to create a hybrid energy system (HES) that integrates various power sources, such as fuel cells and solar photovoltaic (PV), with the existing utility grid, thereby satisfying energy needs while minimizing dependency on conventional fuel-based energy sources like coal and oil. To achieve this, a thorough examination of the energy demand, availability of renewable resources, and current power infrastructure is conducted. This examination focuses on optimizing the design of the HES by considering critical factors such as grid integration, power generation capacity, energy storage capacity, and control strategies. The feasibility and performance of the proposed HES are assessed using a combination of simulation tools, mathematical modeling, and system analysis methodologies. The study carefully evaluates key factors such as system efficiency, reliability, and cost-effectiveness to ensure a durable and economically viable solution. The abstract of this study emphasizes the main quantitative findings, such as a 25% decrease in energy costs and a 30% boost in overall system efficiency, positioning the HES as an attractive choice for sustainable energy management. In addition to the technical aspects, the paper examines the environmental impacts of the HES, particularly its contribution to reducing carbon emissions and promoting clean energy usage. The research seeks to enhance the sustainability and efficiency of the city's energy supply by reducing reliance on fossil fuels, paving the way for a transition to more resilient and sustainable energy solutions. The findings underscore the potential benefits of incorporating renewable energy resources into the existing system, which could lead to lower greenhouse gas emissions, increased energy independence, and improved energy security for the city or facility.

Health care is a critical point for cloud computing, where it offers affordable and optimal data management. Nevertheless, some challenges continue to arise regarding medical information security especially in the cloud. In the light of this, we put forward a Revolutionary Cloud Secure Healthcare Framework. Data anonymization, hybrid encryption, optimal key selection and blockchain integration technologies are easily integrated. This full framework is grounded with a sound foundation for secure medical data interchange, and it protects patient’s privacy. Thus, in our method, data aggregation and encryption preparation under the differential privacy-based anonymization chosen for Electronic Health Record (EHR) data are performed in a careful manner. Asikey encryption algorithm and key-agreement are being used; The Advanced Secure Elliptic Encryption (ASEE) model combines AES-256 with ECIES for powerful encryption, while the Self-Adaptive Wildebeest Herd Optimization (SA –WHO) method ensures secure key selection. The Ethereum blockchain allows both sensitive and non-sensitive EHR data to be securely encrypted as it is transmitted. A rigorous evaluation validates its effectiveness in the secure transfer of medical data in cloud-based healthcare. Metrics confirm effectiveness of the framework based on comparison with industry standards. This Cloud-based Secure Healthcare Framework guarantees unprecedented security and confidentiality for medical data in the cloud, which increases trust among healthcare stakeholders.

Generating electricity near the consumption can provide more flexibility to supply various services to consumers as well as reduce system losses. The limited fossil fuels and air pollution are among the main incentives for the expansion of this technology. One of the perspectives proposed for effective increase in these resources’ involvement is to combine these resources with the objectives of the visibility of the relationship between distributed generation resources and power network, as well as control of these resources more efficiently. One of the methods of combining distributed generation sources is a new concept called micro grid; a topic that is of great importance in this regard is the proper management of distributed generation resources in micro grid aimed to reduce the costs of generating electricity and polluting the environment. The resource management in micro grid is a completely non-linear and top-order problem. In this paper, micro grid optimum programming has been studied considering the capacities available in the electricity market. In order to realize this, as well as reduce pollution and cost simultaneously, NSGAII multi-objective genetic algorithm is used. Studies have been conducted in the form of three scenarios on a sample micro grid consisting of solar, wind, micro turbine, fuel cell and battery resources, considering the uncertainty of load and solar and wind generations. The probabilistic distribution function (PDF) and Roulette Wheel were used to create variables with uncertainty. The first scenario objective was to reduce costs by PSO, GA, and ABC algorithms’ optimization. In the second scenario, pollution was reduced and the studies were repeated using these three algorithms. Finally, in the third scenario, there was a simultaneous reduction in pollution and cost by these three algorithms, taking into account weight coefficients as well as NSGAII algorithm.

Alzheimer's Disease (AD) is a progressive neurodegenerative disorder characterized by memory loss, behavioral changes, and impaired self-care, often preceded by Mild Cognitive Impairment (MCI). Not all MCI cases progress to AD, creating a diagnostic challenge. This study proposes a novel framework for early AD diagnosis using T1-weighted Magnetic Resonance Imaging (MRI). The approach integrates the Adaptive Moving Self-Organizing Map (AMSOM), a neural network technique for unsupervised training and tissue segmentation, with K-means clustering and Principal Component Analysis (PCA) for feature selection. AMSOM dynamically updates neuron weights to improve segmentation accuracy. Classification is performed using various algorithms, evaluated on sensitivity, accuracy, precision, and similarity metrics. Compared to existing techniques such as Fuzzy C-means (FCM) and hybrid Self-Organizing Mapping-K-means (SOM-FKM), the proposed method demonstrates statistically significant improvements in tissue segmentation and classification. It achieved a mean accuracy of 99.8%, reducing the Mean Squared Error (MSE) from 2.3 to 0.44 and improving the Discriminative Overlap Index (DOI) and Tissue Clarity (TC) values to 0.435105 and 0.282381, respectively. Implemented in MATLAB, this method provides a robust, efficient framework for early AD detection, surpassing existing approaches in precision and reliability.