ServDeg: Service Degradation Annotated Network Flow Datasets

Five daily network flow datasets (MON–FRI) with service degradation (SD) annotations, collected in a university dormitory network and used in the IEEE Access paper "Unveiling Latency-Induced Service Degradation: A Methodological Approach With Dataset" (DOI: 10.1109/ACCESS.2024.3456588). Each file is an Apache Parquet dataset of bidirectional network flows (~3.2 million flows per day, 110 columns) comprising NFStream-derived flow features (packet/byte counts, packet size and inter-arrival time statistics, TCP flag counters, SPLT sequences), LAN delay/jitter measurements, and service degradation annotations (SDd, SDj, SD event sequences and timestamps). IP addresses and MAC-derived identifiers are anonymized. The accompanying analysis notebooks and documentation are available in the GitHub repository: github.com/FlowFrontiers/ServDeg-Dataset. Storage note: relative to the originally distributed files, these Parquet files are losslessly recompressed (Zstandard level 22, dictionary/delta/byte-stream-split encodings) and integer columns are downcast to the smallest type that holds each column's value range. All values are bit-for-bit identical to the original data; only the declared integer widths and the compression codec differ. See the included README.md for details.