| ---
|
| license: cc-by-nc-4.0
|
| task_categories:
|
| - tabular-classification
|
| - time-series-forecasting
|
| tags:
|
| - cybersecurity
|
| - phishing
|
| - bec
|
| - email-security
|
| - social-engineering
|
| - threat-intelligence
|
| - apt
|
| - synthetic-data
|
| - awareness-training
|
| - mfa-bypass
|
| pretty_name: CYB004 — Synthetic Phishing Campaign Dataset (Sample)
|
| size_categories:
|
| - 1K<n<10K
|
| ---
|
|
|
| # CYB004 — Synthetic Phishing Campaign Dataset (Sample)
|
|
|
| **XpertSystems.ai Synthetic Data Platform · SKU: CYB004-SAMPLE · Version 1.0.0**
|
|
|
| This is a **free preview** of the full **CYB004 — Synthetic Phishing Campaign
|
| Dataset** product. It contains roughly **1 / 48th of the full dataset** at
|
| identical schema, actor-tier distribution, and statistical fingerprint, so
|
| you can evaluate fit before licensing the full product.
|
|
|
| > 🤖 **Trained baseline available:**
|
| > [**xpertsystems/cyb004-baseline-classifier**](https://huggingface.co/xpertsystems/cyb004-baseline-classifier)
|
| > — XGBoost + PyTorch MLP for 7-class phishing campaign-phase prediction,
|
| > group-aware split by campaign, multi-seed evaluation (ROC-AUC 0.937 ± 0.010),
|
| > honest disclosure of which tasks need the full dataset.
|
|
|
| | File | Rows (sample) | Rows (full) | Description |
|
| |-------------------------------|---------------|---------------|---------------------------------------------------|
|
| | `victim_topology.csv` | ~32 | ~2,400 | Victim organization departments + gateway configs |
|
| | `campaign_summary.csv` | ~100 | ~4,800 | Per-campaign aggregate outcomes |
|
| | `campaign_events.csv` | ~1,254 | ~48,000 | Discrete campaign event log |
|
| | `campaign_trajectories.csv` | ~3,952 | ~280,000 | Per-timestep campaign trajectory data |
|
|
|
| ## Dataset Summary
|
|
|
| CYB004 simulates end-to-end phishing campaign lifecycles across enterprise
|
| email gateways with calibrated user-behavior and detection outcomes, covering:
|
|
|
| - **4 actor capability tiers**: opportunistic, organized_crime, targeted,
|
| nation_state_apt — with per-tier lure personalization, stealth weights,
|
| and BEC attempt probabilities
|
| - **Gateway architectures**: legacy signature, ML classifier, sandbox
|
| detonation, ensemble layered, with per-architecture detection strength
|
| - **Awareness training levels**: none, annual, semi-annual, quarterly,
|
| monthly — with empirically-calibrated CTR reduction factors
|
| - **8 department types**: finance, hr, engineering, executive, sales,
|
| legal, operations, it_admin — each with distinct base CTR and
|
| credential-submission propensity
|
| - **9 campaign phases**: reconnaissance, infrastructure_setup, lure_creation,
|
| send_wave, gateway_evaluation, user_interaction, credential_capture,
|
| lateral_pivot, exfiltration
|
| - **BEC, MFA bypass (AiTM), and lateral phishing** outcome paths
|
| - **DMARC enforcement** modeling with adoption × efficacy factors
|
|
|
| ## Trained Baseline Available
|
|
|
| A working baseline classifier trained on this sample is published at
|
| **[xpertsystems/cyb004-baseline-classifier](https://huggingface.co/xpertsystems/cyb004-baseline-classifier)**.
|
|
|
| | Component | Detail |
|
| |---|---|
|
| | Task | 7-class phishing campaign-phase classification |
|
| | Models | XGBoost (`model_xgb.json`) + PyTorch MLP (`model_mlp.safetensors`) |
|
| | Features | 53 (after one-hot encoding); pipeline included as `feature_engineering.py` |
|
| | Split | **Group-aware by campaign_id** — train/val/test campaigns disjoint |
|
| | Validation | Single seed + multi-seed aggregate across 10 seeds |
|
| | Demo | `inference_example.ipynb` — end-to-end copy-paste |
|
| | Headline metrics | XGBoost: accuracy 0.649 ± 0.038, macro ROC-AUC 0.937 ± 0.010 (multi-seed) |
|
|
|
| The model card documents an honest finding worth knowing before licensing:
|
| **actor-tier attribution is below majority baseline on the sample's 100
|
| campaigns** once leakage-prone features are removed (a sample-size and
|
| per-campaign-constant-feature constraint, not a method failure — the
|
| full 335k-row dataset has ~4,800 campaigns and supports tier attribution
|
| properly). The baseline pivots to **campaign-phase prediction**, which
|
| is strongly learnable on the sample data (ROC-AUC 0.94, stable across
|
| 10 seeds) and is itself a real email-security use case for SOAR
|
| playbooks and threat-hunting workflows.
|
|
|
| ## Calibrated Benchmark Targets
|
|
|
| The full product is calibrated to 12 benchmark metrics drawn from
|
| authoritative sources (Proofpoint State of the Phish, KnowBe4 Industry
|
| Benchmark, Cofense PIQ, Mandiant M-Trends, FBI IC3 BEC Report, Verizon DBIR,
|
| CISA Advisories, APWG eCrime). The sample preserves the same calibration:
|
|
|
| | Test | Target | Observed | Verdict |
|
| |------|--------|----------|---------|
|
| | click_through_rate_fleet | 0.0970 | 0.0873 | ✓ PASS |
|
| | credential_submission_rate | 0.0620 | 0.0562 | ✓ PASS |
|
| | gateway_block_rate_ensemble | 0.8650 | 0.8648 | ✓ PASS |
|
| | bec_attempt_rate | 0.0850 | 0.0850 | ✓ PASS |
|
| | user_report_rate | 0.2100 | 0.1991 | ✓ PASS |
|
| | lure_personalisation_score_apt | 0.7800 | 0.7872 | ✓ PASS |
|
| | mfa_bypass_success_rate | 0.1580 | 0.1584 | ✓ PASS |
|
| | lateral_phishing_pivot_rate | 0.0720 | 0.0700 | ✓ PASS |
|
| | campaign_success_rate | 0.3400 | 0.3000 | ✓ PASS |
|
| | dmarc_enforcement_block_rate | 0.3100 | 0.3108 | ✓ PASS |
|
| | actor_attribution_risk_score | 0.2550 | 0.2958 | ✓ PASS |
|
| | awareness_training_effect_delta | 0.4200 | 0.4200 | ✓ PASS |
|
|
|
| *Note: some benchmarks (e.g. campaign success rate) require larger sample
|
| sizes to converge tightly. The full product passes all 12 benchmarks at
|
| Grade A- or better.*
|
|
|
| ## Schema Highlights
|
|
|
| ### `campaign_trajectories.csv` (primary file, per-timestep)
|
|
|
| | Column | Type | Description |
|
| |---------------------------------|---------|----------------------------------------------|
|
| | campaign_id | string | Unique campaign identifier |
|
| | actor_id | string | Threat actor ID |
|
| | timestep | int | Step in campaign lifecycle (0–59) |
|
| | campaign_phase | string | 1 of 9 phases |
|
| | actor_capability_tier | string | opportunistic / organized_crime / targeted / nation_state_apt |
|
| | target_department | string | Victim department |
|
| | gateway_architecture | string | Gateway protection layer |
|
| | awareness_training_level | string | none / annual / semi_annual / quarterly / monthly |
|
| | send_volume | int | Emails sent at this step |
|
| | gateway_blocked | int | Whether gateway intercepted |
|
| | emails_delivered | int | Emails reaching inbox |
|
| | click_through_rate | float | Per-step CTR |
|
| | credential_submission_rate | float | Per-step credential capture rate |
|
| | user_report_count | int | User-reported phish at this step |
|
| | mfa_bypass_attempted | int | Whether AiTM/MFA bypass attempted |
|
| | bec_attempt | int | Whether BEC lure was used |
|
| | lateral_pivot_attempted | int | Whether lateral phishing was launched |
|
| | lure_personalisation_score | float | Lure quality score (0–1) |
|
| | operational_stealth_score | float | Action stealth weight (0–1) |
|
| | dmarc_enforcement_active | int | Boolean — DMARC reject policy applied |
|
|
|
| ### `campaign_summary.csv` (per-campaign outcome)
|
|
|
| | Column | Type | Description |
|
| |---------------------------------|---------|----------------------------------------------|
|
| | campaign_id, actor_id | string | Identifiers |
|
| | actor_capability_tier | string | Tier classification target |
|
| | campaign_success_flag | int | Boolean — successful campaign |
|
| | click_through_rate | float | Campaign-level CTR |
|
| | credential_submission_rate | float | Campaign-level CSR |
|
| | gateway_block_rate | float | Fraction of waves blocked |
|
| | user_report_rate | float | Fraction of users reporting |
|
| | bec_attempted_flag | int | Boolean — BEC lure used |
|
| | mfa_bypass_success_flag | int | Boolean — MFA bypass succeeded |
|
| | lateral_phishing_count | int | Count of lateral pivots |
|
| | operational_stealth_score | float | Mean campaign stealth |
|
| | lure_personalisation_score | float | Mean lure personalization |
|
|
|
| See `campaign_events.csv` and `victim_topology.csv` for the discrete event
|
| log and victim organization schemas respectively.
|
|
|
| ## Suggested Use Cases
|
|
|
| - Training **phishing campaign-phase classifiers** —
|
| [worked example available](https://huggingface.co/xpertsystems/cyb004-baseline-classifier)
|
| - Training **phishing detection** models on calibrated per-step behavioral
|
| features (CTR, CSR, gateway block, user report)
|
| - **Actor attribution** modelling — 4-tier classification with realistic
|
| stealth and personalization distributions (full dataset recommended for
|
| adequate per-tier sample size)
|
| - **Gateway architecture benchmarking** — legacy vs ML vs sandbox vs ensemble
|
| - **BEC detection** modelling (FBI IC3-calibrated attempt rates)
|
| - **MFA bypass / AiTM attack** detection benchmarking
|
| - **Lateral phishing** propagation modelling (Mandiant-calibrated pivot rates)
|
| - **Awareness training ROI** simulation (KnowBe4-calibrated reduction)
|
| - **DMARC enforcement effect** analysis (adoption × efficacy modelling)
|
|
|
| ## Loading the Data
|
|
|
| ```python
|
| import pandas as pd
|
|
|
| trajectories = pd.read_csv("campaign_trajectories.csv")
|
| summaries = pd.read_csv("campaign_summary.csv")
|
| events = pd.read_csv("campaign_events.csv")
|
| topology = pd.read_csv("victim_topology.csv")
|
|
|
| # Join trajectory data with campaign-level labels
|
| enriched = trajectories.merge(summaries, on="campaign_id", how="left",
|
| suffixes=("", "_summary"))
|
|
|
| # Actor-tier classification target
|
| y_tier = summaries["actor_capability_tier"]
|
|
|
| # Campaign-level binary outcome
|
| y_success = summaries["campaign_success_flag"]
|
|
|
| # Per-step credential-capture target
|
| y_credstuffed = (trajectories["credential_submission_rate"] > 0).astype(int)
|
| ```
|
|
|
| For a worked end-to-end example with campaign-phase classification,
|
| group-aware splitting, and feature engineering, see the inference notebook
|
| in the [baseline classifier repo](https://huggingface.co/xpertsystems/cyb004-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 CYB004 dataset includes **~335,000 rows** across all four files,
|
| with calibrated benchmark validation against 12 metrics drawn from
|
| authoritative email security and threat intelligence sources.
|
|
|
| 📧 **pradeep@xpertsystems.ai**
|
| 🌐 **https://xpertsystems.ai**
|
|
|
| ## Citation
|
|
|
| ```bibtex
|
| @dataset{xpertsystems_cyb004_sample_2026,
|
| title = {CYB004: Synthetic Phishing Campaign Dataset (Sample)},
|
| author = {XpertSystems.ai},
|
| year = {2026},
|
| url = {https://huggingface.co/datasets/xpertsystems/cyb004-sample}
|
| }
|
| ```
|
|
|
| ## Generation Details
|
|
|
| - Generator version : 1.0.0
|
| - Random seed : 42
|
| - Generated : 2026-05-16 13:56:58 UTC
|
| - Campaign model : 9-phase phishing lifecycle state machine
|
| - Overall benchmark : 100.0 / 100 (grade A+)
|
|
|
