FACTORS ASSOCIATED WITH TRUST IN DEEP-LEARNING-BASED THREAT DETECTION FOR ENCRYPTED DARKNET TRAFFIC

Encrypted traffic and anonymity-oriented networks create challenges for cybersecurity monitoring because the content of communications may not be directly inspected. This study examines factors associated with trust in deep-learning-based, real-time threat detection for encrypted darknet traffic, focusing on privacy issues, threat experience, and encryption knowledge. A mixed-method design was used, combining questionnaire responses from 151 participants with seven semi-structured interviews. The questionnaire demonstrated high overall internal consistency (Cronbach’s α = 0.914). Pearson correlations showed positive associations between Detection Trust and Privacy Issues (r = 0.547), Threat Experience (r = 0.673), and Encryption Knowledge (r = 0.682), with reported p-values below .001. A multiple regression model was statistically significant, F(3,147) = 70.192, p < .001, and explained 58.9% of the variance in Detection Trust (R² = 0.589; adjusted R² = 0.581). Standardised coefficients reported in the thesis were β = 0.187 for Privacy Issues (p = .004), β = 0.341 for Threat Experience (p < .001), and β = 0.377 for Encryption Knowledge (p < .001). Interview analysis identified six themes concerning encrypted-traffic visibility, limitations of conventional tools, the potential of metadata-based deep learning, privacy and ethics, trust and explainability, and real-time implementation requirements. The findings concern participants’ perceptions and statistical associations; the study did not experimentally evaluate a deployed deep-learning detector or establish causal effects. The results indicate that trust-oriented system design should address privacy safeguards, transparent explanations, realistic validation, low latency, and continuous performance monitoring.