Abstract As digital payment ecosystems expand globally, financial institutions increasingly rely on advanced machine learning algorithms to detect cyber-enabled fraud. However, optimizing feature engineering to minimize false positives remains a critical operational challenge. This study empirically evaluates the... As digital payment ecosystems expand globally, financial institutions increasingly rely on advanced machine learning algorithms to detect cyber-enabled fraud. However, optimizing feature engineering to minimize false positives remains a critical operational challenge. This study empirically evaluates the predictive efficacy of static device security signals and absolute transaction amounts in identifying fraudulent financial transactions. Utilizing a quantitative, cross-sectional design, we analyzed the Financial Transaction Intelligence Dataset 2026, comprising 10,000 transactions across 1,500 devices and 773 customers. The methodological approach incorporated descriptive statistics, cross-tabulation, Chi-square tests, and nonparametric distribution comparisons to rigorously assess device telemetry and monetary variables across multiple analytical domains. The findings reveal that classic device-security flags including trusted status, rooted or jailbroken states, and biometric enablement exhibit statistically insignificant associations with actual fraud outcomes in this population. Furthermore, transaction amounts demonstrate a near-uniform distribution lacking discriminative power, with fraud rates remaining remarkably flat across all monetary bins and showing near-zero correlation with critical behavioral risk covariates. These results provide actionable insights for risk analysts developing cloud-hosted analytics platforms. Ultimately, this research establishes that absolute transaction amounts and static device attributes are weak, orthogonal predictors of fraud. To enhance operational efficiency and detection accuracy, financial institutions must pivot away from static, rule based thresholds and instead prioritize dynamic behavioral and contextual features within advanced multivariate machine learning models. Read More Read Less

Abstract The deployment of Terahertz (THz) communications in dense Internet of Things (IoT) networks promises unprecedented data rates for next generation wireless systems, yet this technological advancement is critically undermined by the exceptional susceptibility of THz frequencies to... The deployment of Terahertz (THz) communications in dense Internet of Things (IoT) networks promises unprecedented data rates for next generation wireless systems, yet this technological advancement is critically undermined by the exceptional susceptibility of THz frequencies to dynamic weather induced impairments. This study presents a comprehensive empirical evaluation of weather aware anomaly detection frameworks for cloud-assisted THz-enabled IoT networks, comparing the performance of Isolation Forest against a Deep Autoencoder in identifying multivariate degradation patterns caused by adverse atmospheric conditions. Using a synthetically generated dataset comprising environmental, network communication, and blockchainbased trust metrics, unsupervised models were trained exclusively on Clear weather operational data and evaluated against progressively severe weather conditions including Rain, Storm, and Extreme scenarios. The experimental results demonstrate that the Deep Autoencoder significantly outperforms the Isolation Forest, achieving superior ROC-AUC (0.934 vs. 0.883), Average Precision (0.904 vs. 0.842), and F1-score (0.88 vs. 0.81). Notably, reconstruction error exhibited a near fivefold increase under Extreme weather conditions, confirming progressive network degradation as atmospheric severity intensifies. Furthermore, the analysis reveals cascading effects whereby physical layer weather disruptions propagate upward to degrade blockchain consensus latency and trust scores, threatening decentralized security mechanisms. These findings establish that nonlinear representation learning offers superior sensitivity to subtle, compound degradations inherent in weather affected THz-IoT environments, positioning the Deep Autoencoder as the recommended framework for proactive fault diagnosis and predictive network management in future cloudassisted IoT deployments. Read More Read Less

Abstract The deployment of Large Language Models (LLMs) as autonomous agents in next generation infrastructures, such as Agentic AI-Native 6G networks, introduces severe latency, memory, and operational cost bottlenecks. Existing optimization approaches frequently treat model compression and system... The deployment of Large Language Models (LLMs) as autonomous agents in next generation infrastructures, such as Agentic AI-Native 6G networks, introduces severe latency, memory, and operational cost bottlenecks. Existing optimization approaches frequently treat model compression and system orchestration in isolation, failing to adequately address the multidimensional constraints of modern production environments. This paper presents a systematic, multi criteria taxonomy of LLM inference optimization, bridging granular model level compression techniques with macro level API-layer orchestration strategies. We conduct a rigorous comparative analysis of post training quantization algorithms specifically highlighting activationaware methods like AWQ alongside GPTQ and GGUF evaluating their trade offs across memory efficiency, inference throughput, and predictive quality. Our findings reveal that deployment efficiency is not solely dictated by numerical precision but is fundamentally a co-design problem requiring optimized hardware kernels and adaptive, importance aware parameter protection. Furthermore, we demonstrate that systemlevel interventions, such as intelligent model routing, prompt caching, and deterministic decoupling, yield immediate latency reductions without altering underlying model weights. Ultimately, this study provides a comprehensive, scenario based decision framework for matching specific quantization algorithms and APIlayer strategies to distinct deployment environments, ranging from high throughput GPU serving to resourceconstrained edge devices and safety critical autonomous systems. By unifying these perspectives, we establish a foundational guide for practitioners aiming to build reliable, high performance inference pipelines that ensure scalable and economically sustainable AI integration across diverse industrial domains. Read More Read Less