Year: 2026
Type: article
Abstract: Recent years, mitigating bot threats has become a challenging task. In addition to the enormous impact of malicious activities perpetrated by bots, the growth in internet usage has contributed significantly to the current situation. Damage to IT infrastructure, economic losses, human user dissatisfaction in certain service delivery environments, among other activities, are directly associated with malicious bots. The problem becomes even more complex because, on some occasions, human users use mobile applications with their user accounts to gain access privileges to certain commercial services. In other words, the level of sophistication of bots is increasingly high, which means that, under certain circumstances, human activity patterns exhibit the same characteristics as bot activity. With this level of development, detection tasks become increasingly complex and vital. In this study, we propose a meta-learning-based detection approach that supports detection through the characterization of user data (both bots and humans). The characterization process is based on a multi-classifier built from data from previous episodes, which used a Proactive Forest-based classifier. Statistical analysis is performed to select the most appropriate multi-classifier (Bagging, Boosting, Voting, or Stacking). Performance, measured by the percentage of correctly characterized instances, showed that the Voting multi-classifier performed best, with an average of 99.6% of correctly characterized instances.
Source: Revista Angolana de Ciencias
DOI: 10.54580/r0801.14
Author: Hélder João Chissingui
Institution: Instituto Superior Técnico Militar
Language: English
Topic: Spam and Phishing Detection
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7166317699