بحوث سكوبس — هدى وصفي حسون الدليمي

Particularly in handling sensitive data, the growing dependence on artificial intelligence (AI) for data analytics poses serious security problems. Conventional encryption methods hinder safe AI adoption by difficulty balancing privacy with computational efficiency. This paper presents a Confidential AI framework enabling privacy-preserving data analytics by merging homomorphic encryption (HE) with deep learning. To guarantee end-to- end confidentiality, the suggested method lets encrypted data be handled straight without decryption. We create a new homomorphic encrypted deep neural network (HEDNN) minimising latency and computational overhead by optimising encrypted computations. Experimental assessments show that, on encrypted datasets, our framework achieves 85% inference accuracy while preserving a 30% decrease in processing overhead over current HE-based AI systems.These results display how feasible safe AI-driven analytics are without sacrificing performance or accuracy. Where information privateness is important, the suggested paradigm unearths tremendous makes use of in cybersecurity, banking, and healthcare. By advancing secure synthetic intelligence techniques, our paintings helps to create dependable AI systems that hold statistics confidentiality in sensible settings. © 2025 IEEE.

In security camera systems, early fire detection is crucial as prompt alerts might significantly enhance safety measures, preventing property damage and preserving lives. Traditional fire detection systems could be slow to react and create numerous false alarms as they rely on human characteristics and are sensitive to environmental changes. This study uses Deep Convolutional Neural Networks (CNNs) to offer a real-time fire detection system that precisely learns and detects fire patterns. A deep convolutional neural network model is trained using this method on a huge dataset, including photographs of flames and non-fires taken under various lighting and weather conditions. It depends mostly on optimization and data augmentation techniques to enhance the model and lower its susceptibility to overfitting. Among the main outcomes demonstrating the effectiveness of the proposed system in real-time applications are a respectable 96.8% detection accuracy, a low false alarm rate, and an average detection time of < 0.6 seconds per frame. The model outperforms more contemporary and traditional deep learning methods in speed and reliability. At last, the proposed approach generates early alerts and monitors fires in real-time using deep convolutional neural networks (CNNs), which provide a consistent and fast method. © 2025 IEEE.

As cities become larger and people's requirements evolve, traditional approaches to providing urban services have become a challenge in satisfying those requirements. The SmartCityNext method has been developed to address the growing issues with these systems, thereby making them more efficient and effective. This study aims to investigate how Artificial Intelligence (AI) and Machine Learning (ML) can revolutionize the cities' operations, enabling them to be more data-driven, responsive, and effective. The study's primary objective is to identify effective strategies for enhancing urban management systems. A combination of methods has been utilized in several key service sectors, including traffic management, waste collection, energy delivery, and emergency response. The objective has been fulfilled by combining real-time data collected from Internet of Things devices with analytics based on AI and predictive machine learning techniques. The prototype models were tested in several cities, and their results suggested that their implementation led to more effective resource utilization and accelerated service delivery. One of the most significant findings was that the time it took to offer service dropped significantly, and operational efficiency rose by as much as 35%. Additionally, predictive models enabled the performance of preventive Maintenance and improved routing, resulting in significant savings on electricity and gasoline. The study's results indicate that artificial intelligence and machine learning can be effectively utilized in city government, allowing for adaptation as required. These approaches may accelerate the process of generating new ideas that are beneficial for people, cost-effective, and environmentally friendly in future smart cities. © 2025 IEEE.

The need for real-time, user-friendly communication interfaces has driven gesture recognition system development. This need spurred system growth. This need has pushed Human-Computer Interaction (HCI) development forward. A hybrid deep learning architecture called GestureNet-HCI extracts spatial data using CNNs and LSTM networks to mimic temporal sequences. GestureNet-HCI integrates these neural networks. This study would benefit from our suggested structure. The technology is designed for real-time applications and accurately detects static and dynamic hand motions. The system had this feature. This was meant to streamline the procedure. It can collect and examine gesture data from depth video sources and RGB. Using video frames, the CNN module encodes visual-spatial characteristics. The LSTM, the other hand, learns motion continuity and temporal links. It is its responsibility. Resilience is attained by the model using an efficient sliding window frame sampling method and a data augmentation pipeline. You do this to become more resilient. Both elements offer the appropriate strength. With little Latency, GestureNet-HCI reaches state-of-the-art accuracy. Experiments with publicly available gesture datasets indicate that it is suitable for real-time uses. Among the pertinent data in these sets are Chalearn LAP IsoGD, NVGesture, and others. The proposed design allows for natural and effective interaction in smart surroundings, assistive technology, and virtual reality. This is a significant advancement. This allows a new standard for smart, responsive human-computer interaction (HCI) systems. © 2025 IEEE.

One must understand social media to utilize it for targeted advertising, crisis management, and information sharing. Current community recognition and influencer identification approaches typically fail when confronted with huge, ever-changing, and heterogeneous social networks. This research introduces a Novel Graph-based Community Detection and Influencer Identification System (GBCDIS) using adaptive network partitioning and multi-level centrality fusion. GBCDIS uses a data-driven Edge Weight Recalibration Module (EWRM) to dynamically change network design according to user interaction intensity and frequency. Using Hybrid Label Propagation and Modularity Optimization (HLPMO), with random walk-based graph traversal to identify communities at scale while retaining structural integrity. A Rank-Weighted Influence Score (RWIS) rates influential people by degree, betweenness, proximity, and time engagement. Studies on Twitter and Facebook datasets show that the system outperforms baseline models in modularity, detection accuracy, and influence precision. Due to its adaptability to network circumstances, the proposed GBCDIS detects influential community members and unveils their hidden structures. This technique offers a flexible, informative framework for impact modeling and advanced social network research. © 2025 IEEE.

As cities develop bigger and more complicated, it's harder to administer them well. Some of the problems with traditional urban systems are slow responses, data that is spread out, and poor use of resources. This study introduces a Real-Time AI-Driven Digital Twin (AI-DT) platform to tackle these issues. Using cutting-edge technologies like artificial intelligence (AI), edge computing, and the Internet of Things (IoT), this technique generates digital twins, which are virtual copies of real metropolitan infrastructure. These twins obtain data in real time and can utilize it to solve problems, find them early, and make smart decisions on their own. Deep learning helps the AIDT system detect patterns, reinforcement learning helps systems develop over time, and graph-based AI helps it understand how different parts of a city are linked. It also uses federated learning to keep data private while letting different city systems share insights. Tests with data from big cities showed that the AI-DT technology made the system more accurate, faster, and more dependable in general. The findings indicate that this platform can assist smart cities in enhancing their emergency response, traffic management systems, and energy consumption. This study signifies a substantial progression in the creation of more intelligent, swift, and efficient urban management systems via artificial intelligence. © 2025 IEEE.