Accurate prediction of product lifecycle stages is crucial for enhancing inventory turnover and strategic planning in the tobacco industry. This paper proposes an intelligent prediction model that integrates Convolutional Neural Networks (CNN) and Gated Recurrent Units (GRU), further optimized by an Improved Grey Wolf Optimizer (IGWO). The model fuses multi-source enterprise data—including sales trends, IoT logistics information, environmental conditions, and inventory records—to dynamically forecast lifecycle stages and remaining durations. The dataset comprises 180,000 labeled samples collected from real-world tobacco enterprise operations, encompassing multi-source variables such as sales volume, inventory changes, logistics routes, and environmental feedback. Experimental evaluations based on this dataset demonstrate that the proposed IGWO-CNN-GRU model achieves a Mean Squared Error (MSE) of 2.13, a Mean Absolute Error (MAE) of 1.17, and an R² of 0.932, significantly outperforming baseline models. In practical deployment simulations, the prediction deviation is limited to ±5 days, improving allocation efficiency and reducing inventory risks. The approach provides a robust and adaptable solution for full-lifecycle management in tobacco supply chains, offering practical value for intelligent production and market deployment strategies.
This research proposes an efficient parallel graph partitioning algorithm for the big data environment, aiming to solve the bottlenecks of traditional clustering techniques in terms of processing speed and scalability. The algorithm adopts a multi-level graph partitioning framework, decomposing the network information processing task into multiple levels, gradually simplifying the graph structure and backtracking refinement, thereby significantly reducing the computational complexity while ensuring the partitioning quality. The algorithm focuses on balancing the node cohesion within partitions and the edge cutting cost of inter-partition communication. By constructing a global objective function, it minimizes the number of edges across partitions and the workload differences among various sub-graphs, thereby achieving a more balanced partitioning result. The research results show that this algorithm achieves a resource utilization rate of 0.95. In the Hadoop cluster environment, 95% of the computing resources are effectively used for actual task processing, which is significantly higher than that of the competing algorithms. The energy efficiency ratio reaches 0.98, indicating that the number of tasks completed per unit of energy consumption is close to the optimal level, which is superior to the 0.78 to 0.67 range of existing methods, reflecting the advantages of this algorithm in green computing. The load imbalance rate is only 0.00395, and the point weight imbalance rate is 0.00141, which are much lower values than those of the comparison algorithm. This indicates that the algorithm achieves a high degree of balance in task allocation and node weight distribution, effectively avoiding resource waste and performance bottlenecks.
To solve the problems that the weights and thresholds of discrete Hopfield neural networks are easy to fall into local optima and have insufficient anti-noise ability in digit recognition, a digit recognition method based on discrete Hopfield neural network is proposed, which is optimized by fish swarm algorithm and called AFSA-HOP integration method. The parameters of the discrete Hopfield neural network are optimized by using AFSA's powerful global search ability, and the recognition accuracy of the Hopfield neural network is taken as the fitness function. This allows the Hopfield neural network to maintain a high associative success rate even under high noise-to-signal ratios. Computer simulation experiments show that while the recognition performance of the traditional Hopfield neural network significantly deteriorates when the noise intensity is 0.2, the AFSA-HOP method maintains a high recognition accuracy even at noise intensities of 0.4 and 0.5, demonstrating superior digital recognition performance. This method provides a robust new approach for digital recognition and could be further extended in future applications by integrating other optimization algorithms.
This paper presents a novel real-time fire detection framework tailored for IoT devices by integrating the fine-tuned YOLOv10 model with the Accelerator module. Trained on the FireSmokeDataset (Roboflow) and an additional dataset we collected via Roboflow, the system covers fire, smoke, and distracting objects. Optimized for resource-constrained edge devices, the framework demonstrates exceptional performance, achieving high mean average precision (mAP) for fire and smoke detection, with metrics exceeding 84% and a maximum mAP50 of over 91%. We target deployments in residential homes, industrial facilities, and forest monitoring stations. A key contribution of the proposed framework is the construction of a diverse dataset encompassing fire, smoke, and distracting objects - an element often overlooked in existing fire detection datasets. Additionally, fine-tuning the YOLOv10 model components in conjunction with hardware acceleration ensures both prediction accuracy and improved inference response performance. Comprehensive evaluations confirm the system's robustness, scalability, and practicality under various operating conditions. Through experimental analysis, the YOLOv10-S (small) model stands out for its balance between efficiency and resource usage, making it a suitable choice for low-cost real-time applications with resource constraints. By utilizing the Coral Accelerator, the proposed framework reduces inference time by 58% compared to CPU-based implementations, achieving a latency of just 1.7 seconds per frame. The system's lightweight design ensures reliable deployment in remote areas with limited computational resources and unstable network connectivity, maintaining high accuracy while minimizing false alarms.
The September 2025 (Vol. 33, No. 3) issue of CIT. Journal of Computing and Information Technology brings four papers from the areas of graph data processing, computer vision, and business intelligence.
Cloud computing is an environment where resources are utilized on a pay-per-use basis. As users request services, resources are allocated to their associated tasks to ensure timely completion. The cloud service provider aims to optimize resource allocation for these tasks in order to maintain proper utilization of cloud resources. So the proper allocation of task i.e, scheduling of tasks plays a significant role in the performance of the cloud services. This paper offers a comprehensive study of various existing task scheduling algorithms in the cloud environment and an analysis of different scheduling attributes considered in these algorithms.
Network intrusion detection systems are crucial for preserving cybersecurity in the face of increasingly sophisticated cyberthreats that target international networks. Conventional approaches usually use signature-based methods, which are not very good at spotting new attacks like zero-day vulnerabilities. The current study explores machine learning-based methods to enhance NIDS's ability to detect known and unexpected threats. Our classifiers include Gaussian Naive Bayes, Bernoulli Naive Bayes, Logistic Regression, XGB Classifier, and K-Nearest Neighbors Classifier. For training and evaluation, we employ two popular datasets, UNSW-NB15 and NSL-KDD, which provide comprehensive representations of both malicious and benign traffic. Our results suggest that machine learning models may considerably boost intrusion detection's adaptability and accuracy, even though issues like feature selection, data imbalance, and processing efficiency need more research. The foundation of next-generation cybersecurity systems that can adjust to the constantly shifting threat landscape can be machine learning, as this study shows. NIDS can more precisely identify known and unknown attacks by using AIIML methods that learn from historical data to identify patterns and irregularities. This study examines a number of popular machine learning classifiers for intrusion detection using various datasets. The results shows that our study strengthens cybersecurity defense against different intrusion attacks.
As the number of android applications available in the market are increasing rapidly, various types of security attacks using the android apps have also increased at the same pace. The ransomware attack is one of these kinds of security attacks in which the attackers lock the user's phone, encrypts user's data or blocks the user's access to their own data and threatens the user to pay a ransom to gain the access back. Android devices are more prone to these ransomware attacks when compared to Windows and IOS devices. In this work, a novel hybrid ransomware detection method is proposed that analyses image data, text, and application code to extract plain or encrypted threat text. Threatening text is a potential tool and could be one of the most effective features for ransomware detection. Our approach to ransomware detection combines static and dynamic analysis, along with multiple machine learning models acting as classifiers. Notably, our experiments show promising results with an accuracy rate reaching up to 91% and a low rate of false negatives.
In our paper, we have tried to implement some enhancements to the B+ tree structure with the aim to improve its performance and throughput. Our focus is on improving the performance of B+ tree on SIMD architecture machines. By referencing prior research, we identified significant gaps between traditional B+ tree implementations and the capabilities of SIMD architectures, and we have tried to implement some solutions to overcome those gaps. These gaps primarily stem from the mismatch in how data is processed and accessed between the two paradigms. Traditional B+ trees are optimized for sequential data access patterns, whereas SIMD architectures excel in parallel data processing. We have tried to identify the disparities and implement some solutions to overcome them. Our approach aims to mitigate the adverse effects of these gaps, thereby enhancing the B+ tree's efficiency in modern computing environments. Through empirical evaluation and comparative analysis, we demonstrate the potential improvements in performance and throughput achievable with these modifications.
Traditional crime analysis methods are overwhelmed by the ever-growing volume and intricacies of crime data. This research addresses this challenge by proposing a novel system utilizing machine learning to analyze crime data in India. Focusing on “Serious Fraud - Cheating” and “Serious Fraud - Criminal Breach of Trust” across all states, the system tackles the limitations of manual analysis. By integrating k-means clustering and Decision Trees, the system efficiently processes and interprets complex data, achieving high accuracy and precision in identifying patterns. This paves the way for proactive policing strategies, empowering law enforcement to anticipate and prevent future crimes. As technology continues to reshape criminal activity, this system offers a valuable tool for law enforcement agencies to stay ahead of tech-savvy criminals by embracing cutting-edge analytical approaches.
The Sustainable Development Goals (SDGs) call for high-quality data to inform policy design, progress tracking, and impact evaluation. Geospatial data is crucial for measuring targets and monitoring SDGs, as it studies the earth's processes like climate change, carbon fluxes, water dynamics, and biodiversity. Geospatial data helps detect social and economic inequalities by visualizing regional differences. The present paper demonstrates GeoNIRF model which is the Geospatial Cloud Computing enabled National Institutional Ranking Framework (NIRF) for data storage and visualization of the top 100 Indian Universities as per the NIRF 2023 report to achieve the aim of SDG #4. The GeoNIRF model focuses on SDG#4, the quality of education including the most current advances in cloud computing and Geospatial technology. It also develops a geospatial database on which Web Feature Services can be applied. Finally, it derives the qualitative assessment of the different aspects of the state-wise ranking using geospatial data to achieve SDG#4.
Federated Learning (FL) is decentralized machine learning, which preserves data confidentiality and security but suffers challenges such as high communication overhead, latency, and scalability issues in large scale smart city networks. We propose a Hierarchical Federated Learning (HFL) framework which takes advantage of fog nodes to address these problems. HFL framework ensures that the cost of communication is cut back through the introduction of the multilevel aggregation strategy where local models are aggregated first at fog nodes before they are combined at a central server. Communication costs are reduced while latency is improved and scalability is enhanced by this HFL framework through simulations on real-world smart city datasets. On edge devices, our simulation results with real-world datasets from smart cities show that compared to traditional FL it reduces communication overhead by up to 50%, achieves faster model convergence with similar accuracy, and leads to lower energy consumption. This framework represents a big step forward towards deployment of FL in smart cities making it efficient and scalable in resource-constrained environments.
Fog computing using Unmanned Aerial vehicles is one of the most emerging technologies in the modern era. It has various advantages like high mobility, increased scalability, wireless connectivity, minimum man-power requirement, etc. Unmanned Aerial vehicles can act as fog devices between the Cloud and Internet of Things remote applications. It enhances the connectivity for the IoT devices available in remote areas by transferring the data to the cloud server in less time. In this paper, we propose algorithms that could be followed for the proper and optimal data transmission to the cloud from IoT remote applications via fog Nodes.
This paper introduces the design and prototype of a new home automation system that utilizes GSM technology as the network infrastructure to connect its components. The proposed system is composed of two primary parts: the first is the GSM module, which acts as the core of the system, managing, controlling, and monitoring the user's home. Users and system administrators can connect to the GSM locally to access devices and manage system functions. The second part is the hardware interface module, which provides the necessary interface for relays and actuators within the home automation system. The mobile phone, originally designed for making calls and sending text messages, has evolved into a versatile device, especially with the advent of smartphones. In this study, the researcher develops a home automation system using GSM and Arduino, allowing users to control household appliances by simply sending SMS commands through their GSM-based phones. This paper states that a smartphone is not necessary; but an old GSM phone can effectively be used to turn home electronic appliances on and off from any location. The proposed system offers greater scalability and flexibility compared to commercially available home automation systems.
This paper introduces PaperChain, a blockchain-based platform that addresses the inefficiencies, security vulnerabilities, and outdated processes inherent in traditional document management systems. Traditional systems struggle with issues like fraud, data breaches, and manual errors. PaperChain tackles these challenges with a decentralized approach using Filecoin's infrastructure and the Interplanetary Consensus (IPC) system. The platform provides a tamper-proof record-keeping environment, leveraging smart contracts for automation and PaperChain cryptocurrency (PRC) for transactions. The integration of MetaMask further simplifies authentication, while Filecoin's decentralized storage ensures advanced cryptographic validation mechanisms to protect against document tampering and duplication. PaperChain sets a new standard for secure and transparent document management in the digital age.
In recent years, microservice architecture has garnered substantial traction, highlighting the critical role of resilience patterns in ensuring robust system performance under adverse conditions. Among the principal techniques for failure management within microservices is the circuit breaker pattern, a vital component for maintaining system integrity. Enhancing its resilience continues to be a significant challenge. Various methodologies have been proposed to optimize circuit breakers by tweaking different parameters. However, the open state interval parameter—during which dependent services await the recovery of the disrupted service—has traditionally remained static. This work tackles this issue by introducing a dynamic adjustment mechanism for the open-state interval of circuit breakers. The empirical results indicate that this dynamic adjustment significantly boosts the number of successful requests by optimizing resource allocation, enhancing throughput, and mitigating cascading failures across service replicas. By adapting the open state interval in response to real-time conditions, the system can better manage variable loads and failures, thereby improving overall performance and stability in microservice architectures.
Load balancing techniques are vital for the efficient allocation of workloads across nodes in cloud infrastructures, playing an indispensable role in optimizing overall performance. By ensuring that load factors are managed effectively, these techniques significantly enhance system efficiency. This is particularly relevant in data centers, which serve as fundamental elements within cloud computing frameworks. Effective utilization of data centers can lead to marked improvements in system performance. A key aspect of successful load balancing is the precise computation of workloads among nodes. To address this, numerous load balancing models have been developed, particularly tailored for public cloud environments. These models often employ partitioning strategies to optimize performance. This paper delves into existing partitioning concepts and load balancing models within the realm of public clouds. Furthermore, it introduces a comprehensive generalized model designed to adeptly handle a variety of network load scenarios, thereby advancing the field of load balancing in cloud computing.
Online clothing purchasing has become a widespread practice for millions of individuals worldwide in the last several years. It has advanced significantly and is now a regular activity for many customers. The use of virtual try-on technology, which allows customers to view how particular things look on them prior to making a purchase, is what is facilitating this advancement. Shopping online is gaining more and more popular, especially for clothes. Nevertheless, the inability to physically try on clothing when shopping online limits customers' ability to visualize how a garment might look on them. The technology can produce high-fidelity try-on photos that maintain the features and general look of clothing items. Apart from the target garment region, the method further inserts the face, hair, lower garments, and legs to preserve model's personality characteristics. The two modules that comprise the system are the Try-On Module (TOM) and the Geometric Matching Module (GMM). In the Geometric Matching Module, in-store clothing items are warped to the intended image of a person with great precision using grid interval consistency loss and an occlusion management technique. The transformation is regularized by grid interval consistency loss to prevent distortion of clothing patterns. Additionally, an occlusion handling technique promotes appropriate warping even when target bodies are obscured by arms or hair. Then, the face, hair, lower clothing, and legs are added to the model's representation for the Try on Module in order to preserve personality characteristics other than the target clothing region.
In the data-driven era, applications that generate vast amounts of data have become a central focus. The exponential growth in data generation poses significant challenges in data analysis. Financial transactions, in particular, have become increasingly complex, necessitating effective methods for detecting anomalies. Unnoticed irregularities can lead to substantial problems for banks and other financial institutions, including financial losses and eroded trust. In many real-world applications, dealing with imbalanced data is a critical concern. While most classification methods focus on two-class data problems, addressing a solution for class-imbalanced scenarios is equally essential. This work proposes a methodology that applies the SMOTE algorithm to various Machine Learning (ML) and Deep Learning (DL) models, aiming to balance the imbalanced data and improve classification performance. In the practical comparison of various ML and DL models with and without the SMOTE technique, this work also experimentally examines and then discusses the challenges in identifying fraudulent transactions over two different financial transactional datasets. Finally, by using SOMTE with various ML and DL models, this work presents 37% - 91 % improvement on the banking transactional dataset and 57% −98% improvement on the online shopping transactional dataset in terms of F1-Score.
Identifying vehicles across multiple non-overlapping camera views is a crucial task in computer vision, with important applications in public safety, urban surveillance, and intelligent transportation systems. Traditional methods are prone to errors under variations in scale, illumination, complex background, and object occlusion scenarios. To address these limitations, we propose a new framework, Coarse and Fine-grained Vision Transformer (CFViT). This framework efficiently captures coarse-grained(global) and fine-grained (local) features. We then develop a Coarse Fine Grained Module (CFGM) that performs shuffle and shift operations on vehicle patch images, resulting in robust feature representation ensuring robustness against variations in vehicle appearance. We further encode side information of vehicles into the Faster Vision Transformer (FasterViT) model to prevent failures in distinguishing the same object from different viewpoints. To validate the effectiveness of the proposed method, we achieve state-of-the-art performance on two publicly available datasets, including VehicleID, and Veri-776.
