ADDRESSING ANOMALY DETECTION CHALLENGES IN AI-ENABLED MOBILE AND SMART NETWORKS

Abstract

Ubiquitous connectivity, powered by mobile networks and smart digital technologies, istransforming critical sectors such as public health, industrial automation, and smart cities. However, the increasing complexity and interdependence of these connected sys- tems present unprecedented challenges in maintaining performance, security, and reliabil- ity, particularly for mission-critical applications that demand ultra-reliable, low-latency communication. Traditional anomaly detection systems, which rely on reactive fault re- porting, are no longer sufficient, as they often identify issues only after service degradation has occurred. This dissertation addresses these limitations through four novel AI-driven anomaly de- tection frameworks that enable proactive, scalable, and context-aware intelligence by integrating emerging mobile networks, user mobility traces, and secure pervasive connec- tivity facilitated by smart devices. First, a hybridHealth AI framework is introduced as a smartphone-based screening tool that combines on-device acoustic analysis for rapid de- tection with deep learning-based detailed diagnostics in cloud environments, optimizing resource efficiency and ensuring high diagnostic accuracy and fairness, while addressing challenges such as data inconsistencies, label scarcity, and demographic biases. To ad- vance screening beyond the individual level and enable population-scale epidemic hotspot prediction, EpiTrack is proposed as a novel outbreakprediction framework that leverages real-time user mobility traces from mobile networks to enable accurate and proactive public health interventions. This data-driven machine learning model offers unprece- dented forecasting precision, helping prevent outbreaks from escalating into widespread epidemics. In addition, smart-device-based screening tools and mobility-driven analytics rely heavily on continuous mobile network connectivity; any disruption directly undermines real-time anomaly detection and predictive intelligence. To maintain system reliability in antici- pation of such disruptions, we propose a proactive anomaly management framework that models network reliability, predicts faults in advance, optimizes resource allocation, and supports zero-touch automation in 6G and beyond networks. An additional barrier to achieving resilient, autonomous AI-driven systems lies in the vulnerabilities introduced by compromised or untrustworthy data collected through mobile networks and smart dig- ital technologies. To address this challenge, we propose MRIF, a multimodule AI-based security framework that identifies and mitigates data-driven attacks, ensures accurate mobility insights, safeguards optimization processes, and reinforces trust in network in- telligence. By addressing the key challenges of fairness, proactivity, reliability, and security, this dissertation establish a unified, end-to-end ecosystem for AI-driven anomaly detection. Through this convergence of intelligent modeling and real-world integration, the work paves the way for next-generation autonomous systems that are resilient, secure, and significantly impact the public health outcomes worldwide.