Journal of Networking Technology H. Ketmaneechairat, P. Oothongsap, A. Mingkhwan 17 4 2026 https://doi.org/10.6025/jnt/2026/17/4/222-239 https://www.dline.info/jnt/fulltext/v17n4/jntv17n4_3.pdf The increasing deployment of federated learning, edge computing, and Internet of Things (IoT) infrastructures necessitates robust frameworks for evaluating data transmission reliability in secure environments. Traditionally, communication quality is assessed using classical network performance metrics such as bandwidth, latency, packet loss, encryption time, and distributed node count. However, it remains unclear whether these conventional indicators adequately capture continuous transmission reliability under complex security, privacy, and trust constraints. This study investigates the predictive utility of five classical network metrics across two distinct outcomes: continuous transmission reliability and categorical transmission status. Using a Network Transmission Dataset comprising 13,000 records, we applied correlation analysis, linear and non-linear regression (Ordinary Least Squares, Random Forest, Gradient Boosting), and classification models (Multinomial Logistic Regression, Random Forest). The results reveal a striking dissociation: classical metrics exhibit near zero correlation with continuous reliability (R2  , failing to explain its variance. Conversely, the same metrics achieve approximately 72% accuracy in classifying categorical transmission status, driven predominantly by packet loss (54.6%) and network latency (30.9%). These findings demonstrate that classical network metrics are fundamentally insufficient for predicting continuous transmission reliability in federated secure environments. Comprehensive reliability assessment frameworks must therefore incorporate multidimensional security, privacy, and trust variables rather than relying exclusively on traditional Quality of Service (QoS) indicators.