SynNetQoS: A Transparent Simulation-Based Synthetic 4G/5G Dataset Generator for QoS and QoE Modeling

​Public mobile-network datasets that jointly expose session-level Quality of Service (QoS), Quality of Experience (QoE), radio, mobility, device, traffic-demand, and service-state variables remain limited because operational traces are often restricted by privacy, commercial sensitivity, and incomplete reproducibility. This paper presents SynNetQoS, a transparent simulation-based synthetic data-generation framework for controlled 4G/5G QoS/QoE modeling. The reference release contains 50{,}000 time-correlated records from 5{,}000 sessions and 67 columns spanning radio, mobility, device, traffic demand, QoS, QoE, and service labels. The generator encodes explicit assumptions rather than learning from restricted operator traces. It combines a selected 3GPP TR 38.901-inspired UMi-Street Canyon large-scale propagation layer with LOS/NLOS assignment, carrier-frequency-aware path loss, shadow fading, contextual signal penalties, mobility and handover rules, congestion and tower-load dynamics, application-specific demand, heuristic streaming-QoE scoring, and probabilistic dropped-connection labeling. Reproducibility is supported by fixed configuration, deterministic scripts, audit fields, saved metadata, integrity checks, Monte Carlo summaries, and a dataset SHA-256 hash. SynNetQoS is evaluated as a bounded synthetic research artifact, not a replacement for operator traces, field measurements, ns-3, 5G-LENA, or packet-level simulation. External alignment uses selected comparable variables from Vienna phone measurements and Campus QoS data, focusing on RSRP and download throughput. A controlled KPI-level 5G-LENA/ns-3 trace comparison examines offered-load throughput trends. Leakage-aware supervised-learning benchmarks with session-wise splits assess structured learnability across QoS/QoE impairment and future dropped-connection tasks. The results support SynNetQoS as a reproducible artifact for controlled QoS/QoE modeling and benchmark design.