electricsheepafrica/africa-cyber-patient-safety

African Cyber Patient Safety Dataset

Dataset Description

Synthetic dataset of cybersecurity incidents with direct patient safety impact in African healthcare, paired with secure baseline records. Cyber incidents in healthcare can cause direct patient harm through device malfunction, treatment delays, misdiagnosis, and medication errors. This dataset focuses on these safety-critical outcomes.

Research Gap Addressed

Patient safety research focuses on clinical errors, not cyber-induced patient harm. The ECRI Institute and FDA track cybersecurity threats to patient safety, but African contexts are absent. African healthcare faces additional patient safety cyber risks: power-dependent devices with no UPS, manual workaround errors during system downtime, and delayed referrals due to network outages.

African Healthcare Context

African cyber-patient safety scenarios: ransomware forcing reversion to paper records with transcription errors, infusion pump crashes during power outages, misdiagnosis from corrupted PACS images, medication errors from EMR unavailability, cancelled surgeries from ransomware, and patient diversion from emergency departments. Limited backup power means cyber incidents compound with infrastructure failures.

Intelligence Sources

This dataset is parameterised from verified real-world intelligence:

Source	URL
FDA Cybersecurity and Patient Safety Guidance	https://www.fda.gov/medical-devices/digital-health-center-excellence/cybersecurity-medical-devices
HIMSS Patient Safety Organization Report	https://www.himss.org/resources/patient-safety
WHO Patient Safety Framework	https://www.who.int/teams/integrated-health-services/patient-safety
ECRI Top 10 Health Technology Hazards 2024	https://www.ecri.org/resources/health-technology-hazards

Dataset Summary

• Size: 10,000 rows (5,000 attack incidents + 5,000 secure baseline records)
• Features: 50+ columns including facility metadata, attack characteristics, security posture indicators, impact metrics, and response outcomes
• Target: label (1 = security incident, 0 = secure baseline)
• Countries: 20 African nations with weighted distribution reflecting digital economy and healthcare incident prevalence
• Intelligence Traceability: Every attack record includes intelligence_source and intelligence_source_url fields
• Synthetic Flag: All records include is_synthetic = 1 for transparency
• Seed: 42 for full reproducibility

Supported Tasks

• Tabular Classification: Predict security incidents vs. secure baselines
• Anomaly Detection: Identify outlier patterns in healthcare security posture
• Risk Scoring: Train models to compute composite security risk scores
• Feature Engineering: Domain-specific composite scores included (see Engineered Features below)
• Benchmarking: Cross-dataset comparison of African healthcare cybersecurity risks

Data Acquisition

All records are synthetically generated using weighted-choice parameterisation calibrated against verified real-world intelligence reports. The generation process:

1. Intelligence Source Mapping: Each attack type and pattern is derived from documented incidents in the source reports
2. African Contextualisation: Country weights, facility types, and infrastructure characteristics reflect African healthcare realities
3. Balanced Sampling: Equal representation of attack and benign classes (5,000 each)
4. Leakage Prevention: Build pipeline filters out duplicate or near-duplicate records across attack and benign sets
5. Feature Extraction: Domain-specific composite scores are computed post-generation

Key Columns

Identification & Source:

• record_id — Unique identifier
• intelligence_source — Organisation that reported the real-world incident this record is parameterised from
• intelligence_source_url — URL to the source report
• country — African nation
• is_synthetic — Always 1 (transparency flag)

Core Incident Features: incident_type, system_affected, patients_affected, harm_type, severity_level, treatment_delay_hours, mortality_risk, device_malfunction

Engineered Features (computed by features.py):

Feature	Description
patient_harm_severity	Composite of harm type, severity, and mortality risk
impact_scale_score	Number of patients affected weighted by severity
mortality_flag	Binary for incidents with mortality risk >= high
incident_criticality	Overall criticality combining system type, harm, and patient count

Usage

Load from Hugging Face

from datasets import load_dataset

dataset = load_dataset("electricsheepafrica/africa-cyber-patient-safety", split="train")
print(dataset.features)
print(dataset["label"][:10])  # 1 = incident, 0 = baseline

Pandas

import pandas as pd

df = pd.read_csv("data/processed/cyber-patient-safety_africa_dataset.csv")
X = df.drop(["label", "record_id", "intelligence_source", "intelligence_source_url"], axis=1, errors="ignore")
y = df["label"]

Quick Baseline Classifier

from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split

numeric_df = df.select_dtypes(include=["number"]).drop(["label", "is_synthetic"], axis=1, errors="ignore")
X_train, X_test, y_train, y_test = train_test_split(numeric_df, df["label"], test_size=0.2, random_state=42, stratify=df["label"])

clf = RandomForestClassifier(n_estimators=200, random_state=42, n_jobs=-1)
clf.fit(X_train, y_train)
print(f"Accuracy: {clf.score(X_test, y_test):.3f}")

Data Splits

Single train split provided. Users should perform stratified train/validation/test splits:

from sklearn.model_selection import StratifiedKFold
cv = StratifiedKFold(n_splits=5, shuffle=True, random_state=42)

Limitations

• Synthetic Data: All records are synthetically generated and do not represent real individual patients or incidents
• Simplified Dynamics: Real-world attack chains are more complex than the feature representation
• Static Snapshot: Does not capture temporal evolution of threats or seasonal patterns
• Generalisation: Models trained on this data should be validated on real-world African healthcare security data before deployment
• Ethical Considerations: While synthetic, the dataset simulates sensitive healthcare security scenarios and should be used responsibly

Citation

@dataset{cyber_patient_safety,
  author = {Electric Sheep Africa},
  title = {African Cyber Patient Safety Dataset},
  year = {2026},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/electricsheepafrica/africa-cyber-patient-safety},
  type = {Synthetic Dataset}
}

License

CC BY 4.0

Contact

Electric Sheep Africa