Abstract
One of the most important issues in modern analytics is how to make financial forecasts more accurate when the data is insufficient or inconsistent. When data is sparse, normal imputation approaches don't work as well and don't pick up on nuanced relationship information. This paper proposes a machine learning-based method to consolidate financial data from several sources and execute generalized missing value imputation. Using regularized tensor completion, this method successfully models nonlinear and high-dimensional financial structures. We learn strong and noise-resistant embeddings by using an orthogonal regularization method to avoid overfitting in sparse settings. This article talks about RegDeepTensor, an architecture for deep neural networks. It combines MLP with structural regularization, embedding learning, and modeling nonlinear interactions. The solution lets you break down multi-source financial datasets quickly and supports both single and coupled tensor factorization. This keeps the accuracy of the forecast high while reducing down on the amount of work that needs to be done. Tests on real-world applications, such figuring out how good a bond is and how much money a company makes per share, reveal that it works better than the other options. The suggested method works far better than current linear and nonlinear tensor completion models, with improvements of up to 74%. It also makes embedding better by cutting down on noise and redundancy in learned representations. The results show that combining regularization with deep embedding learning makes financial forecasting more accurate and scalable.
Key Words
embedding learning; financial forecasts; machine learning; RegDeepTensor
Address
Balamohan S.: Department of CSE, Bharath Institute of Higher Education and Research(BIHER), Chennai, Tamil Nadu, India
V. Khanaa: Department of IT, Bharath Institute of Higher Education and Research (BIHER), Chennai, TamilNadu, India
Abstract
The high rate of smart ecosystem development has accelerated the need to manage resources effectively and intelligently in IoT settings where conventional rule-based frameworks and current machine learning or optimization methods can be challenged by the lack of scalability, high latency, and responsiveness to dynamic and heterogeneous data flows. The existing methods, such as regular deep learning and classical optimization, are not effective to represent contextual dependencies and guarantee optimal resource allocation. The paper aims to overcome these constraints by presenting a smart IoT system based on AI and incorporating two new sequential algorithms, the Deep Attribute Context Convoluted Network (DACCN) and the Cyber Crow Resource Optimization Algorithm (CCROA). The DACCN, which is an AI-based algorithm, uses deep contextual attribute-driven convolutional learning to model complex spatial-temporal relationships and make true resource demand predictions, which is then used as input by the CCROA, which is an optimization-based algorithm based on swarm intelligence, which dynamically allocates resources by minimizing allocation error and optimizing system performance under varying constraints. The proposed model is tested on the publicly available Smart* Home Dataset (UMass Trace Repository), and the results of the experiment indicate that the model has greatly enhanced the accuracy of prediction, resource utilization, and response time as compared to the existing baseline methods, which prove the usefulness of the model in managing the intelligent resources in smart ecosystems.
Key Words
cyber crow resource optimization algorithm; deep attribute context convoluted network; intelligent allocation; IoT resource management; smart ecosystems; z-normalization
Address
Mohd Hasham Ali, Ahmed Abdul Muneem: Department of Mechanical Engineering Department, Muffakham Jah College of Engineering and Technology, Hyderabad, Telangana, India
Afshan Kaleem: Department of Electronics and Communication Engineering Department, Muffakham Jah College of Engineering and Technology, Hyderabad, Telangana, India
Sreelatha Mutyala: Department of Information Science and Engineering Department, Gopalan College of Engineering and Management, Bangalore, India
S. Irfan Sadaq: Department of Mechanical Engineering Department, Muffakham Jah College of Engineering and Technology, Hyderabad, Telangana, India
Abstract
In modern software testing environments, the rapid evolution of application data and system behavior demands adaptive and intelligent test case generation methods. This work presents a Cognitive Optimization framework for dynamic data enhancement in test case framing, designed to improve testing efficiency and coverage under changing data conditions. The proposed model integrates cognitive computing principles with optimization strategies to analyse real-time input variations and automatically refine test cases. By employing a self-learning mechanism, the system continuously evaluates test data relevance, prioritizes critical inputs, and eliminates redundancy, resulting in an optimized and adaptive test suite. The algorithm enhances the accuracy and performance of the testing process by identifying hidden data patterns and maintaining consistency in evolving datasets. Experimental evaluation demonstrates that the proposed cognitive optimization approach achieves higher fault detection rates and reduced execution time compared to conventional static testing techniques. This makes it an effective and intelligent solution for next-generation automated software testing environments.
Key Words
ensemble algorithm; machine learning; software testing; test case optimization
Address
Manojkumar V.: Department of Computer Applications, Vels Institute of Science, Technology and Advanced Studies (VISTAS), Chennai, Tamil Nadu 60011, India
R. Mahalakshmi: Department of Advanced Computing and Analytics, Vels Institute of Science, Technology and Advanced Studies (VISTAS), Chennai, Tamil Nadu 60011, India
Abstract
The fileless execution, anti-forensic evasion, and dynamic nature of memory resident malware are a significant challenge to modern digital forensics as the malware forms cannot be detected using traditional disk-based detection. To overcome these difficulties, this paper presents a smart, multi-stage, memory malware forensics with a Deep Quantum Gated Multi-Fusion Neural Network (DQGMF-NN) trained with the Bio-Scaled Spider Optimization (ANT-BSSOA). The steps start with acquiring live system RAM to obtain memory dumps, and then integrating Volatility and Rekall to obtain memory artifacts parsing and object extraction. To reduce the noise, entropy-based, page de-duplication, ASLR normalization, and cross-view consistency checking, Adaptive memory preprocessing is achieved by a Z-Score Min-Max Resilient Memory normalization (Zs-MMRN) algorithm to enhance the reliability of the memory objects including processes, threads, DLLs, kernel modules, registry entries, and network sockets. Then, the Malware Behavior Pattern Extraction Algorithm (MBPEA) models temporal and structural abnormalities identifying fine-grained signature of behavior through suspicious API sequence, code injection, reflective DLL loading, hidden processes, privilege escalation, and kernel anomalies. Extraction features are then coded or represented with graph and sequence representations in order to maintain spatial and temporal correlations, and are optimized with ANT-BSSOA to eliminate redundancy and choose the most informative feature subsets. Lastly, DQGMF-NN detection engine is involved with hierarchical feature fusion and deep quantum learning to identify and categorize the types of malwares including benign, malicious, APTs, rootkits, and fileless malware. Experimental tests on real-world memory dump datasets show that the proposed framework has good detection accuracy, low false positives, strong real-time functionality and forensic soundness which makes it very useful in detecting and investigating malware that reside in memory.
Abstract
Internet of things (IoT) technologies are now widely used in the healthcare industry. Due to the growing demands of the Internet of Things, enormous amounts of patient data are being collected and used for diagnosis. In order to reliably identify diseases in real-time scenarios, artificial intelligence (AI) approaches have recently advanced. Security, energy constraints, and a lack of training data remain the main problems in the IoT-enabled medical field, despite its advantages. Blockchain technology, a decentralised architecture that is extensively used, was recently developed to achieve security. This study presents a new blockchain based approach that enables secure medical data transmission and diagnosis through distributed ledger technology. The proposed model aims to diagnose the illness with the highest possible detection rate while safely transmitting medical images. This model integrates several phases of the process, including image capture, blockchain, encryption, and diagnosis. After that, the diagnostic procedure includes neural network-based classification, segmentation based on U-Net model, and DenseNet based feature extraction. Using benchmark medical images, the experimental performance of the proposed technique was verified, and the obtained results demonstrated the technique's increased performance. While the feature extraction is based on DenseNet, the suggested model achieved maximum classification performance with sensitivity of 97.94%, specificity of 98.56%, and accuracy of 98.92%.
Key Words
artificial intelligence; blockchain; deep learning; internet of things; medical image analysis
Address
M. Kavitha: Department of Information Technology, Karpagam Institute of Technology, Coimbatore, Tamil Nadu, India
J. Nirmaladevi: Information Technology, KGiSL Institute of Technology, Coimbatore Tamil Nadu, India
A. Chandrasekar: Department of Computer Science and Engineering, Nandha College of Technology, Erode, Tamil Nadu, India
Sahana Vasudev: Research and Development Department, Takeda Pharmaceuticals, Boston, Massachusetts, USA
