Scopus Indexed Publications

Paper Details


Title
Predicting WordPress Security Levels from Malware and Vulnerability Signals: An Ordinal-Hierarchical Machine Learning Approach with XAI

Author
Md. Kausaruzzaman, Md. Sadekur Rahman,

Email

Abstract

WordPress is one of the most commonly used content management systems, and, as such, a prime source of cyberattacks due to unreliable use of third-party tools, insecure setups, weak authentication features, and weaknesses caused by exposure. This paper introduces an ordinal-hierarchical machine learning model to predict the level of security of WordPress when based on malware, vulnerability and exposure features. The experimental data set contains 331 real-world compromised WordPress websites, each of which is characterized by 20 binary security indicators and three-ranked security classes, namely Low, Medium and High. To better represent the multidimensional nature of the cybersecurity risk, the starting indicators were coded into semantically relevant groups and converted into hierarchical attributes with count-based metrics, normalized risk ratings, presence indicators, and terms of interaction. The prediction problem was represented as an ordinal classification problem that would be decomposed into two binary classifiers through threshold based decomposition, and then probability would be rebuilt to give the overall security-level prediction. Ten machine learning models were evaluated comparatively and it was revealed that Gradient Boosting overall had the highest level of performance with its accuracy of 68.66% and a macro F1-score of 0.6771. In addition, explainable artificial intelligence was integrated with SHAP to make the model more interpretable and find out which factors have the most significant impact on estimating the level of security. The results indicate that the proposed framework offers a structured and interpretable approach to organized WordPress security-level assessment. At the same time, the moderate accuracy shows that this remains a challenging prediction task and that the framework is best used to support, rather than replace, expert security triage.


Keywords

Journal or Conference Name
2026 6th International Conference on Emerging Smart Technologies and Applications, eSmarTA 2026

Publication Year
2026

Indexing
scopus