--- license: cc-by-nc-4.0 task_categories: - tabular-classification - time-series-forecasting tags: - cybersecurity - network-traffic - intrusion-detection - synthetic-data - anomaly-detection - apt - c2-beacon pretty_name: CYB001 — Synthetic Network Traffic (Sample) size_categories: - 1K 🤖 **Trained baseline available:** > [**xpertsystems/cyb001-baseline-classifier**](https://huggingface.co/xpertsystems/cyb001-baseline-classifier) > — XGBoost + PyTorch MLP, copy-paste inference notebook, full metrics and > honest limitations in the model card. | File | Rows (sample) | Rows (full) | Description | |-------------------------|---------------|---------------|--------------------------------------------| | `network_topology.csv` | ~200 | ~3,200 | Network segments and defender configs | | `session_summary.csv` | ~1000 | ~62,000 | Multi-flow session aggregates | | `network_flows.csv` | ~9,770 | ~500,000 | Per-flow records (CICFlowMeter-compatible) | | `flow_events.csv` | ~5,431 | ~120,000 | Per-flow security event records | ## Dataset Summary CYB001 simulates 30 days of enterprise network traffic across **9 segment types** (corporate LAN, DMZ, cloud workload, OT/ICS, endpoint fleet, SOC management plane, zero-trust, guest Wi-Fi, data centre spine), with: - **3-class labels**: `BENIGN`, `MALICIOUS`, `AMBIGUOUS` - **19 fine-grained traffic categories** including portscan, brute-force, SQLi, XSS, exfiltration, C2 beaconing, lateral movement, ransomware staging - **4 attacker capability tiers**: opportunistic, targeted, APT, insider threat - **Diurnal traffic patterns** with off-peak attack bias for APT/insider tiers - **APT C2 beacon regularity** governed by a configurable IAT coefficient of variation (default 0.065 → regularity score ≈ 0.93) All IP addresses are SHA-256 pseudonyms (`IP_<12 hex>`) — no real network data. ## Trained Baseline Available A working baseline classifier trained on this sample is published at **[xpertsystems/cyb001-baseline-classifier](https://huggingface.co/xpertsystems/cyb001-baseline-classifier)**. | Component | Detail | |---|---| | Task | 3-class flow classification (`BENIGN` / `MALICIOUS` / `AMBIGUOUS`) | | Models | XGBoost (`model_xgb.json`) + PyTorch MLP (`model_mlp.safetensors`) | | Features | 101 (after one-hot encoding); pipeline included as `feature_engineering.py` | | Demo | `inference_example.ipynb` — end-to-end copy-paste | | Headline metrics | XGBoost test accuracy 0.998, macro-F1 0.996 — synthetic; see model card for limitations | This is a reference baseline, not a production IDS. The model card documents the calibrated signals it picks up, an ablation showing the model is not session-dominated, and six explicit limitations including the gap between synthetic and real-world traffic. ## Calibrated Benchmark Targets The full product is calibrated to 12 benchmark metrics; the sample preserves the same calibration. Observed values on this sample: | Test | Target | Observed | Verdict | |------|--------|----------|---------| | malicious_flow_rate | 0.1720 | 0.2015 | ✓ PASS | | c2_beacon_regularity_score | 0.7800 | 0.7673 | ✓ PASS | | payload_entropy_benign_mean | 4.8000 | 4.8556 | ✓ PASS | | protocol_violation_rate | 0.0150 | 0.0161 | ✓ PASS | | scan_probe_density | 0.0430 | 0.0450 | ✓ PASS | | exfil_volume_ratio | 0.0240 | 0.0153 | ✓ PASS | | retransmission_rate | 0.0380 | 0.0365 | ✓ PASS | | dns_query_rate_anomaly | 0.0620 | 0.0620 | ✓ PASS | | lateral_move_flag_rate | 0.0310 | 0.0340 | ✓ PASS | | session_risk_score_apt | 0.6800 | 0.6803 | ✓ PASS | | fwd_bwd_byte_ratio_benign | 1.3400 | 1.4119 | ✓ PASS | | tunnel_detection_rate | 0.0180 | 0.0180 | ✓ PASS | ## Schema ### `network_flows.csv` (primary file) | Column | Type | Description | |------------------------------|---------|----------------------------------------------| | flow_id | string | Unique flow identifier | | session_id | string | Parent session FK | | source_ip_hash | string | SHA-256 pseudonymised source IP | | destination_ip_hash | string | SHA-256 pseudonymised destination IP | | source_port, dest_port | int | TCP/UDP port numbers | | protocol | string | TCP / UDP / HTTPS / DNS / SMTP / SSH / etc. | | flow_start_timestamp | string | ISO timestamp | | flow_duration_ms | int | Flow duration in milliseconds | | total_fwd_packets | int | Forward packet count | | total_bwd_packets | int | Backward packet count | | total_bytes_fwd | int | Forward byte volume | | total_bytes_bwd | int | Backward byte volume | | fwd_packet_len_mean / _std | int | Forward packet length statistics | | bwd_packet_len_mean / _std | int | Backward packet length statistics | | flow_bytes_per_sec | float | Throughput (bytes/sec) | | flow_packets_per_sec | float | Throughput (packets/sec) | | inter_arrival_time_mean/_std | float | IAT statistics (ms) — key C2 beacon feature | | tcp_flag_{syn,ack,fin,rst,psh,urg}_count | int | TCP flag counts | | flow_lifecycle_phase | string | initiation / handshake / transfer / etc. | | traffic_category | string | 1 of 19 fine-grained categories | | attack_subcategory | string | Attack subcategory (empty for benign) | | **label** | string | **BENIGN / MALICIOUS / AMBIGUOUS** (target) | | segment_id | string | FK to `network_topology.csv` | | source_device_type | string | workstation / server / iot / mobile / cloud | | dest_device_type | string | (same as source) | | attacker_capability_tier | string | opportunistic / targeted / apt / insider | | retransmission_flag | int | TCP retransmission flag | | fragmentation_flag | int | IP fragmentation flag | | protocol_violation_flag | int | Protocol-violation detection flag | See `session_summary.csv`, `network_topology.csv`, and `flow_events.csv` for the complementary aggregate, topology, and per-event schemas. ## Suggested Use Cases - Training and evaluating **network intrusion detection** models - Benchmarking **C2 beacon detection** algorithms (regular-IAT signatures) - **APT behaviour modelling** with off-peak temporal bias - **Multi-class anomaly detection** including the `AMBIGUOUS` class - **Synthetic-vs-real transfer-learning** studies - **Feature engineering** practice with CICFlowMeter-compatible fields ## Loading the Data ```python import pandas as pd flows = pd.read_csv("network_flows.csv") sessions = pd.read_csv("session_summary.csv") topology = pd.read_csv("network_topology.csv") events = pd.read_csv("flow_events.csv") # Join flows to topology to get defender configuration flows_enriched = flows.merge(topology, on="segment_id", how="left") # Binary classification target y = (flows["label"] == "MALICIOUS").astype(int) ``` For a worked end-to-end example including the 3-class classification target, feature engineering, and predictions, see the inference notebook in the [baseline classifier repo](https://huggingface.co/xpertsystems/cyb001-baseline-classifier/blob/main/inference_example.ipynb). ## License This **sample** is released under **CC-BY-NC-4.0** (free for non-commercial research and evaluation). The **full production dataset** is licensed commercially — contact XpertSystems.ai for licensing terms. ## Full Product The full CYB001 dataset includes **~685,000 rows** across all four files, with calibrated A-grade benchmark validation across 12 statistical tests. 📧 **pradeep@xpertsystems.ai** 🌐 **https://xpertsystems.ai** ## Citation ```bibtex @dataset{xpertsystems_cyb001_sample_2026, title = {CYB001: Synthetic Network Traffic Dataset (Sample)}, author = {XpertSystems.ai}, year = {2026}, url = {https://huggingface.co/datasets/xpertsystems/cyb001-sample} } ``` ## Generation Details - Generator version : 1.0.0 - Random seed : 42 - Generated : 2026-05-16 13:23:38 UTC - Simulation window : 30 days - Overall benchmark : 100.0 / 100 (grade A+)