Frequency-Temporal-Aware Network for Privacy-Preserving Healthcare Data Security and Reliable Ransomware Infection Classification

Abstract

The fast deployment of electronic health records, IoMT devices, cloud computing, and connected clinical architecture has resulted in an increasing attack surface on modern healthcare organizations, where reliable ransomware infection classification becomes a necessity. Current ransomware classification techniques depend mostly on deterministic methods of classifications and are not able to consider, at the same time, privacy issues, complicated behavior pattern, and uncertainties in decision-making process. To fill these gaps, we introduce the concept of Frequency-Temporal-Aware Deep Learning framework for Privacy-Preserving Healthcare Data Security and Reliable Ransomware Infection Classification. The proposed technique will include the use of privacy-preserving pre-processing and feature extraction in order to preserve healthcare data confidentiality and extract security-related behavioral patterns. After that, frequency-domain and temporal representations will be introduced to capture ransomware behavioral patterns, communications, and malicious dependencies in healthcare and IoMT environments.A Frequency-Temporal Aware Network (FTAN) model is built to learn the complementary features in spectral and temporal domains to achieve accurate ransomware infection classification. Moreover, Bayesian parameterization is employed to output probabilistic predictions and determine uncertainty in classification, thus increasing the trustworthiness of security decisions. The performance of the proposed framework is experimentally assessed on MedSec-25 and RanSMAP datasets with different behavioral labels unified into the Legitimate and Ransomware categories. According to the experiments conducted, the proposed approach is able to provide accuracy in the range of 97.3%-98.7%, precision from 97.8%-98.5%, recall from 97.6%-98.3%, and F1-score from 97.7%-98.4%. The discriminative capability of the proposed approach is proved by the ROC-AUC analysis, while predictive entropy is used to assess reliability and distinguish between high-confidence and ambiguous security events. Overall, the proposed framework represents an efficient and privacy-aware solution to protect sensitive health data and ensure reliable classification of ransomware infections.