Authors: Rakesh Mehta

Abstract: The rapid escalation of global data traffic, catalyzed by the proliferation of 5G, Internet of Things (IoT) devices, and high-definition streaming services, has rendered traditional network management techniques increasingly obsolete. Packet Flow Classification serves as the foundational mechanism for Quality of Service (QoS) provisioning, resource allocation, and security enforcement. Historically, flow classification relied on port-based analysis or Deep Packet Inspection (DPI); however, the widespread adoption of end-to-end encryption protocols, such as TLS 1.3 and QUIC, alongside dynamic port allocation, has nullified these legacy methods. This review examines the paradigm shift toward Machine Learning (ML) and Deep Learning (DL) models as the primary engines for real-time traffic classification. By focusing on statistical flow features and byte-level patterns rather than plaintext payloads, ML models can identify applications and malicious intent within encrypted tunnels with unprecedented accuracy. We categorize current methodologies, ranging from classical supervised learners like Random Forests to advanced neural architectures, including Convolutional Neural Networks (CNNs) for spatial feature extraction and Recurrent Neural Networks (RNNs) for temporal sequence modeling. Furthermore, the review addresses the critical challenges of real-time processing at line speed, data imbalance in network datasets, and the necessity for Explainable AI (XAI) in network operations. By synthesizing recent academic breakthroughs and industrial implementations, this paper provides a strategic roadmap for building autonomous, "self-driving" networks. The findings suggest that ML-driven packet flow classification significantly enhances network visibility and resilience, providing the cognitive intelligence required to manage the complex, opaque traffic landscapes of the modern digital era.

DOI: https://doi.org/10.5281/zenodo.19491852

Published by: vikaspatanker

Authors: Manoj Tiwari

Abstract: Machine learning has emerged as a transformative force in cybersecurity, enabling predictive defence mechanisms that move beyond traditional reactive strategies. This review explores the evolution, methodologies, and applications of machine learning models in predictive cybersecurity defence. By leveraging large-scale data, these models can detect anomalies, anticipate threats, and automate responses in real time. Techniques such as supervised learning, unsupervised learning, and deep learning have been widely adopted to identify patterns in network traffic, user behaviour, and system logs. Predictive capabilities allow organizations to mitigate risks before attacks occur, reducing financial and operational damage. However, challenges such as adversarial attacks, data imbalance, model interpretability, and scalability persist. This article also highlights emerging trends, including federated learning, explainable AI, and hybrid defence systems that integrate human expertise with machine intelligence. Through a comprehensive analysis, the review emphasizes the need for robust, adaptive, and ethical frameworks to ensure reliable deployment of machine learning in cybersecurity. The findings suggest that while machine learning significantly enhances predictive capabilities, its effectiveness depends on data quality, continuous model updates, and integration with existing security infrastructures.

DOI: https://doi.org/10.5281/zenodo.19491834

Published by: vikaspatanker

Authors: Siti Amina

Abstract: Deep learning-based intrusion detection systems (IDS) have emerged as a transformative approach for securing enterprise networks in the face of increasingly sophisticated cyber threats. Traditional signature-based and rule-based IDS solutions struggle to detect zero-day attacks, polymorphic malware, and advanced persistent threats due to their reliance on predefined patterns. In contrast, deep learning models offer the ability to automatically learn hierarchical feature representations from large-scale network traffic data, enabling improved detection accuracy and adaptability. This review examines the evolution, methodologies, and practical implementation of deep learning-based IDS in enterprise environments. It highlights the role of architectures such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), autoencoders, and transformer-based models in identifying anomalous and malicious activities. The study further explores data preprocessing techniques, feature engineering, and benchmark datasets commonly used for training and evaluation. Key challenges, including data imbalance, model interpretability, computational overhead, and real-time deployment constraints, are critically analyzed. Additionally, the integration of deep learning IDS with emerging technologies such as cloud computing, edge computing, and software-defined networking (SDN) is discussed. The review concludes by outlining future research directions focused on improving scalability, explainability, and resilience against adversarial attacks. Overall, deep learning-based IDS represent a promising paradigm shift in enterprise cybersecurity, offering intelligent, adaptive, and proactive defense mechanisms.

DOI: https://doi.org/10.5281/zenodo.19491764

Published by: vikaspatanker

Authors: Sanduni Fernando

Abstract: The rapid migration of organizational workloads to cloud environments has introduced unprecedented scalability but also significant financial complexity. Cloud billing is often characterized by high-volume, granular data where "anomalies"—unexpected spikes or shifts in spending—can remain undetected for weeks, leading to "cloud sprawl" and budget overruns. Traditional threshold-based monitoring systems often fail in these dynamic environments due to their inability to distinguish between legitimate scaling and genuine waste. This article reviews the shift toward Machine Learning (ML)-centric approaches for cloud cost anomaly detection. By leveraging time-series forecasting, clustering, and deep learning, ML models can learn the "seasonal" rhythms of business operations and flag deviations with high precision. This review explores the architectural foundations of these systems, evaluates supervised versus unsupervised learning paradigms, and discusses the operational challenges of implementing AI-driven FinOps. Ultimately, the integration of ML transforms cost management from a reactive reporting task into a proactive, automated defense mechanism, ensuring operational stability and financial efficiency in modern cloud-native architectures.

DOI: https://doi.org/10.5281/zenodo.19491753

Published by: vikaspatanker

Authors: Ahmad Rizal

Abstract: Insider threats represent one of the most challenging cybersecurity risks, as they originate from individuals with legitimate access to organizational systems and data. Traditional security mechanisms often fail to detect such threats due to their reliance on signature-based or rule-based approaches that lack contextual awareness. Behavioral analytics, powered by machine learning (ML), has emerged as a transformative approach for identifying anomalous patterns indicative of insider misuse, fraud, or sabotage. This review explores the integration of behavioral analytics and ML techniques to enhance insider threat detection capabilities. By leveraging user activity logs, network traffic data, and system interactions, ML models can establish baseline behavioral profiles and identify deviations in real time. The study examines supervised, unsupervised, and hybrid learning approaches, highlighting their effectiveness in detecting both known and unknown threats. Additionally, it discusses feature engineering, data preprocessing, and the role of contextual information in improving detection accuracy. Challenges such as data imbalance, privacy concerns, adversarial behavior, and model interpretability are also critically analyzed. The review further explores emerging trends, including deep learning, graph-based analytics, and explainable AI, which are shaping next-generation insider threat detection systems. Ultimately, behavioral analytics

DOI: https://doi.org/10.5281/zenodo.19491716

Published by: vikaspatanker

Authors: Muhammad Hakim

Abstract: The escalating complexity of global digital infrastructures, characterized by the convergence of 5G, massive IoT deployments, and hyperscale cloud-to-edge continuums, has rendered traditional linear network management models obsolete. At the heart of this complexity lies the network topology—the intricate map of nodes and interconnections that dictates the flow, latency, and resilience of data. This review article explores the paradigm shift toward Graph Analytics for Network Topology Optimization. Unlike traditional tabular data analysis, graph analytics treats the network as a native mathematical graph, where routers, switches, and endpoints are vertices, and the communication links are edges. This relational perspective allows for the discovery of structural properties—such as centrality, community clusters, and bottleneck bottlenecks—that are invisible to classical monitoring. We categorize the core methodologies of graph-driven optimization, including the use of Graph Neural Networks (GNNs) for predictive traffic steering and PageRank-inspired algorithms for identifying critical infrastructure vulnerabilities. The article examines how graph analytics enables "Topological Resilience," allowing networks to autonomously reconfigure their structure in response to failures or shifting demand. Furthermore, the review addresses the critical challenges of processing massive-scale dynamic graphs in real-time, the computational overhead of graph embeddings, and the necessity for explainable graph models in network operations. By synthesizing recent breakthroughs in spectral graph theory and combinatorial optimization, this paper provides a strategic roadmap for building "Self-Optimizing Topologies." The findings suggest that graph analytics is the foundational intelligence required to manage the "Relational Complexity" of the 6G era, ensuring that global networks are not just faster, but fundamentally more robust, efficient, and adaptive.

DOI: https://doi.org/10.5281/zenodo.19491714

Published by: vikaspatanker

Authors: K.Satheesh, S.Chandana, V.Balaji, Sk.Yaseen, V.V.Bhanu Satya Sri

Abstract: The Medical Image Analysis Tool uses Google's Gemini 2.5 Flash multimodal model to analyze X-rays, MRIs, CT scans, and ultrasound images, generating comprehensive diagnostic reports with findings, diagnoses, and patient-friendly explanations. Integrated with DuckDuckGo search, it enables real- time retrieval of medical literature and treatment protocols for evidence-based recommendations. Built with Streamlit and the Agno framework, the tool delivers structured, medically accurate responses in markdown format for healthcare professionals and students. This AI-powered assistant reduces diagnostic uncertainty, empowers data- driven decision-making, and enhances medical image interpretation efficiency.

Published by: vikaspatanker

Authors: Mrs. Kalluri Jaya Sri Sai, Dheeravath Rajender, Bommapala Manideep, Arishe Pramod

Abstract: With the increasing demand for advanced digital security, efficient and scalable real-time monitoring has become essential. Traditional security evaluation methods often rely on manual oversight or delayed reporting, which lacks the immediate and personalized feedback necessary to thwart modern attacks. This project presents an Intelligent System for Real-Time Cyberthreat Information that leverages automated data streaming to evaluate the digital landscape for threats. The proposed system analyzes network logs and global threat feeds for syntax, logic, and patterns of malicious activity, providing instant alerts along with clear threat explanations and suggested mitigation strategies.

DOI:

 

 

Published by: vikaspatanker

Authors: Mr. Chitoor Venkat Rao Ajay Kumar, Amgothu Shivateja, Dhanavath Praveen, Banoth Bhaskar

Abstract: The “Semantic and Contextual Intelligence-Based Court Verdict Prediction” system uses AI to analyze legal case data and predict outcomes. It understands the meaning and context of legal documents using Natural Language Processing. The system studies past judgments, case facts, and legal patterns to make predictions. It provides structured insights that help legal professionals in decision-making. This improves the speed, accuracy, and efficiency of legal analysis.

DOI:

 

 

Published by: vikaspatanker

Authors: Ms. Gyara Monika, Banothu Malsoor, Mendu Balram Sai Abhishek, Mohammed Abdul Sameer

Abstract: Explainable Artificial Intelligence (XAI) is a system that helps humans understand how machine learning models make decisions. Traditional AI models often work like a “black box,” where the output is given without explaining the reason. XAI provides clear explanations for predictions by showing important features, rules, or visual insights. This improves transparency, trust, and fairness in AI systems, especially in critical areas like healthcare, finance, and education. By making AI decisions understandable, XAI helps users and developers detect errors, bias, and improve model performance.

Published by: vikaspatanker