--- license: cc-by-nc-4.0 task_categories: - tabular-classification - time-series-forecasting tags: - cybersecurity - malware - malware-classification - threat-intelligence - apt - ransomware - synthetic-data - edr - sandbox-evasion - polymorphic pretty_name: CYB003 — Synthetic Malware Behaviour & Classification (Sample) size_categories: - 1K 🤖 **Trained baseline available:** > [**xpertsystems/cyb003-baseline-classifier**](https://huggingface.co/xpertsystems/cyb003-baseline-classifier) > — XGBoost + PyTorch MLP for 10-class malware execution-phase prediction, > group-aware split by sample, multi-seed evaluation (accuracy 0.905 ± 0.010), > honest disclosure of which tasks need the full dataset. | File | Rows (sample) | Rows (full) | Description | |----------------------------|---------------|---------------|------------------------------------------------| | `environment_profiles.csv` | ~100 | ~3,200 | Endpoint environment configurations | | `sample_summary.csv` | ~100 | ~5,600 | Per-sample aggregate KPIs | | `execution_events.csv` | ~1,056 | ~60,000 | Discrete malware lifecycle events | | `malware_samples.csv` | ~6,000 | ~280,000 | Per-timestep sample telemetry | ## Dataset Summary CYB003 simulates malware execution lifecycles across endpoint protection stacks with calibrated detection/evasion outcomes, covering: - **9 malware families**: ransomware, trojan, rootkit, worm, spyware, fileless_malware, cryptominer, botnet_agent, dropper - **4 threat-actor tiers**: commodity, crimeware, apt, nation_state — with per-tier sandbox evasion budgets, LotL (Living-off-the-Land) abuse rates, and polymorphic mutation probabilities - **Endpoint protection stacks**: legacy AV, NGAV (ML-based), EDR - **Static PE features**: entropy, packing detection, section anomalies, import hash distributions - **Behavioural telemetry**: process injection, persistence mechanisms, C2 beacon patterns, lateral spread - **Outcome modelling**: AV signature detection, EDR behavioural detection, sandbox evasion success, family attribution confidence ## Trained Baseline Available A working baseline classifier trained on this sample is published at **[xpertsystems/cyb003-baseline-classifier](https://huggingface.co/xpertsystems/cyb003-baseline-classifier)**. | Component | Detail | |---|---| | Task | 10-class malware execution-phase classification | | Models | XGBoost (`model_xgb.json`) + PyTorch MLP (`model_mlp.safetensors`) | | Features | 69 (after one-hot encoding); pipeline included as `feature_engineering.py` | | Split | **Group-aware by sample_id** — train/val/test samples disjoint | | Validation | Single seed + multi-seed aggregate across 10 seeds | | Demo | `inference_example.ipynb` — end-to-end copy-paste | | Headline metrics | XGBoost: accuracy 0.905 ± 0.010, macro ROC-AUC 0.975 ± 0.002 (multi-seed) | The model card documents an honest finding worth knowing before licensing: **malware-family classification is at majority baseline on the sample's 100 samples** (a sample-size constraint, not a method failure — the full 280k-row dataset has ~5,600 samples and supports family classification properly). The baseline pivots to **execution-phase prediction**, which is strongly learnable on the sample data (91% accuracy, ROC-AUC 0.98, stable across 10 seeds) and is itself a real SOC use case for dynamic-analysis and EDR phase tagging. ## Calibrated Benchmark Targets The full product is calibrated to 12 benchmark metrics drawn from authoritative threat intelligence and AV-testing sources (VirusTotal, AV-TEST, MITRE ATT&CK Evaluations, Mandiant M-Trends, CrowdStrike GTR, Verizon DBIR). The sample preserves the same calibration: | Test | Target | Observed | Verdict | |------|--------|----------|---------| | av_detection_rate_commodity | 0.6200 | 0.6319 | ✓ PASS | | edr_detection_rate_apt | 0.3100 | 0.3096 | ✓ PASS | | sandbox_evasion_rate_nation | 0.7200 | 0.7225 | ✓ PASS | | lateral_propagation_rate | 0.0950 | 0.1038 | ✓ PASS | | pe_entropy_mean_packed | 0.9100 | 0.9100 | ✓ PASS | | lotl_abuse_rate_apt | 0.4300 | 0.4300 | ✓ PASS | | dwell_time_ratio_apt | 0.3200 | 0.3198 | ✓ PASS | | family_attribution_confidence | 0.6800 | 0.6808 | ✓ PASS | | c2_detection_rate | 0.5400 | 0.5394 | ✓ PASS | | campaign_success_rate | 0.3400 | 0.2900 | ✓ PASS | | polymorphic_detection_penalty | 0.2400 | 0.2392 | ✓ PASS | | false_negative_rate_fileless | 0.3800 | 0.4203 | ✓ PASS | *Note: some benchmarks (e.g. campaign success rate, lateral propagation) require larger sample sizes to converge tightly. The full product passes all 12 benchmarks at Grade A- or better.* ## Schema Highlights ### `malware_samples.csv` (primary file, per-timestep telemetry) | Column | Type | Description | |------------------------------|---------|----------------------------------------------| | sample_id | string | Unique malware sample identifier | | family_id | string | Malware family instance ID | | actor_id | string | Threat actor ID | | timestep | int | Step in malware lifecycle (0–59) | | malware_family | string | 1 of 9 families | | threat_actor_tier | string | commodity / crimeware / apt / nation_state | | target_platform | string | windows / linux / macos / android | | ep_stack | string | legacy_av / ngav_ml_based / edr_full | | pe_entropy | float | Portable Executable section entropy (0–1) | | packer_detected_flag | int | Whether PE packer was detected | | process_injection_count | int | Process-injection events at this step | | persistence_mechanism | string | registry / scheduled_task / service / wmi | | c2_beacon_active | int | Whether C2 channel is beaconing | | sandbox_evaded | int | Whether sandbox evasion succeeded | | av_detected | int | AV signature detection at this step | | edr_detected | int | EDR behavioural detection at this step | | dwell_time_hours | float | Cumulative dwell time | | lotl_technique_used | string | Living-off-the-Land binary if any | ### `sample_summary.csv` (per-sample outcome) | Column | Type | Description | |-----------------------------------|---------|------------------------------------------| | sample_id, family_id, actor_id | string | Identifiers | | malware_family | string | Family classification target | | threat_actor_tier | string | Tier classification target | | target_platform | string | Platform | | campaign_success_flag | int | Boolean — successful campaign | | av_detection_flag | int | Boolean — AV detection ever | | edr_detection_flag | int | Boolean — EDR detection ever | | sandbox_evaded_flag | int | Boolean — sandbox evasion ever | | packer_detected_flag | int | Boolean — packer detected | | family_attribution_confidence | float | Confidence score (0–1) | | total_dwell_hours | float | End-to-end dwell | | lateral_propagation_count | int | Count of lateral spread events | See `execution_events.csv` and `environment_profiles.csv` for the discrete event log and endpoint environment schemas respectively. ## Suggested Use Cases - Training **malware execution-phase classifiers** — [worked example available](https://huggingface.co/xpertsystems/cyb003-baseline-classifier) - Training **malware family classifiers** (9-class with realistic class imbalance and family-specific feature distributions — full dataset recommended for adequate per-class sample size) - **Threat actor attribution** modelling (4-tier classification) - **EDR detection benchmarking** — packed vs unpacked, signature vs behavioural, fileless vs binary - **Sandbox evasion detection** with tier-calibrated evasion budgets - **Polymorphic malware detection** — sample mutation effects on AV signature coverage - **C2 beacon detection** with realistic beacon-active timestep patterns - **PE entropy / packing detection** — entropy distributions tied to ground-truth packing flags - **Living-off-the-Land binary detection** for APT-tier samples ## Loading the Data ```python import pandas as pd samples = pd.read_csv("malware_samples.csv") summaries = pd.read_csv("sample_summary.csv") events = pd.read_csv("execution_events.csv") environments = pd.read_csv("environment_profiles.csv") # Join per-timestep telemetry with per-sample summary labels enriched = samples.merge(summaries, on="sample_id", how="left", suffixes=("", "_summary")) # Family classification target y_family = summaries["malware_family"] # Threat-actor tier target y_tier = summaries["threat_actor_tier"] # Binary detection target (per-timestep) y_detected = (samples["av_detected"] | samples["edr_detected"]).astype(int) ``` For a worked end-to-end example with execution-phase classification, group-aware splitting, and feature engineering, see the inference notebook in the [baseline classifier repo](https://huggingface.co/xpertsystems/cyb003-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 CYB003 dataset includes **~349,000 rows** across all four files, with calibrated benchmark validation against 12 metrics drawn from authoritative AV-testing and threat intelligence sources. 📧 **pradeep@xpertsystems.ai** 🌐 **https://xpertsystems.ai** ## Citation ```bibtex @dataset{xpertsystems_cyb003_sample_2026, title = {CYB003: Synthetic Malware Behaviour & Classification Dataset (Sample)}, author = {XpertSystems.ai}, year = {2026}, url = {https://huggingface.co/datasets/xpertsystems/cyb003-sample} } ``` ## Generation Details - Generator version : 1.0.0 - Random seed : 42 - Generated : 2026-05-16 13:46:05 UTC - Lifecycle model : Multi-timestep PE + behavioural + outcome simulation - Overall benchmark : 100.0 / 100 (grade A+)