HealthDroid: malware detection in healthcare Android application using BERT

The increase in Android-based interconnected devices poses a great challenge to cybersecurity, most especially because of the fast-rising threat of Android malware. The field of Android malware detection and categorization has escalated over the years, but no work has been done that specifically focuses on health-related applications. The study has conducted an extensive examination that highlights the significant consequences of data breaches in healthcare data that resulted in substantial financial losses for the industry, with estimates reaching $11 million in the year 2023 alone. This paper describes a novel method for better detecting and categorizing Android malware, which specifically focuses on healthcare applications. This study proposes a framework, HealthDroid, that gathers health-related applications using several datasets and the latest methods, such as transfer learning and machine learning algorithms, to make healthcare applications safer. The architecture of HealthDroid integrates Bidirectional Encoder representations from Transformers (BERT) twice. First, as a fine-tuned DistillBERT-NER for the filtration for healthcare-specific applications, and second, ClinicalBERT as a contextual encoder to enrich static code analysis. The domain-specific biomedical BERT ensures healthcare relevance using biomedical corpora such as MACCROBAT, PubMed, MIMIC-III, and hand-curated lexical entities. In this article, researchers also perform several vertical studies to gain deep insights into the impacts of cyberattacks on healthcare organizations. The proposed method outperforms conventional approaches with a 99.47% accuracy rate and fewer false positives.