Anomaly-based threat detection in smart health using machine learning.

Area of Science:

• Computer Science
• Healthcare Informatics

Background:

• Anomaly detection is critical in healthcare due to increasing data complexity and insider threats.
• Electronic Health Records (EHRs) are vulnerable to manipulation, necessitating robust security measures.

Purpose of the Study:

• To propose and evaluate a methodology for securing EHRs against data anomalies using unsupervised machine learning.
• To improve the accuracy and reduce false positives in anomaly detection within healthcare data.

Main Methods:

• Employed a systematic approach including data preprocessing, labeling, and modeling.
• Utilized unsupervised machine learning algorithms: Isolation Forest (IForest) and Local Outlier Factor (LOF) with variations (SVM, Decision Tree, Random Forest).
• Evaluated models using metrics like accuracy, sensitivity, specificity, F1 Score, Silhouette Score, and Dunn Score.

Main Results:

• Isolation Forest SVM achieved the highest accuracy (99.21%), sensitivity (99.75%), specificity (99.32%), and F1 Score (98.72%).
• Isolation Forest Decision Tree also showed strong performance with 98.92% accuracy and 99.35% F1 Score.
• Isolation Forest Random Forest demonstrated lower specificity (72.84%) compared to other models.

Conclusions:

• Isolation Forest SVM is the top-performing model for EHR anomaly detection, offering superior accuracy and reduced false positives.
• The proposed methodology effectively secures sensitive healthcare data against evolving digital threats.
• The approach successfully identified novel contextual anomalies missed by baseline methods.