Predicting User Susceptibility to Phishing Based on Multidimensional Features

Area of Science:

• Cybersecurity
• Human-Computer Interaction
• Machine Learning

Background:

• Phishing remains a significant network security threat, exploiting human vulnerabilities.
• Existing antiphishing techniques require enhancement to address the human element effectively.

Purpose of the Study:

• To develop and validate a multidimensional phishing susceptibility prediction model (MPSPM).
• To identify key features predicting an individual's susceptibility to phishing attacks.

Main Methods:

• Collected data from 1105 volunteers via questionnaires on demographics, personality, knowledge, security behavior, and cognitive processes.
• Constructed legitimate and phishing email datasets for experimental simulation.
• Applied 7 supervised machine learning algorithms to classify users as susceptible or nonsusceptible.

Main Results:

• Machine learning models achieved high accuracy in predicting user phishing susceptibility.
• The maximum prediction accuracy rate reached 89.04% using the proposed multidimensional approach.

Conclusions:

• The multidimensional phishing susceptibility prediction model demonstrates significant potential for identifying at-risk individuals.
• Findings underscore the importance of understanding human factors in cybersecurity and developing targeted interventions.