Hybrid machine learning
Combining complementary statistical and neural approaches for text classification.
RESEARCH & PUBLICATIONS
Machine learning for more trustworthy digital systems.
My research explores hybrid machine learning, email security, high-dimensional classification, and explainable AI. This page brings together published journal articles, ongoing research, and supporting technical outputs, while distinguishing peer-reviewed publications from collaborative and experimental work.
RESEARCH FOCUS
From classification performance to explainable security.
I investigate how machine-learning models can improve the identification of unwanted and potentially malicious messages, and how model outputs can be made more interpretable for people evaluating security decisions.
My work spans hybrid Naïve Bayes–neural-network methods and current experiments combining transformer representations with XGBoost. Explainability methods such as SHAP and LIME are explored to help examine model behaviour and individual predictions.
Email security
Spam and phishing detection, with attention to malicious content and classification errors.
Explainable AI
Using SHAP and LIME to investigate model behaviour and support interpretation.
ONGOING RESEARCH
Transformer–XGBoost for explainable email spam and phishing detection
This research evaluates a hybrid pipeline in which a transformer encodes email text and XGBoost performs downstream classification. The comparative work considers BERT, RoBERTa, and DeBERTa, with SHAP and LIME used to investigate model explanations.
The experimental work draws on public email corpora including Enron, SpamAssassin, and CEAS 2008. Reported results apply to the described experimental subset and evaluation protocol; they should not be interpreted as a guarantee of performance on every live email stream.
PEER-REVIEWED PUBLICATIONS
Published journal articles
Hybrid Machine Learning Algorithms for Email and Malware Spam Filtering: A Review
European Journal of Theoretical and Applied Sciences, 2(2), 76–86
A review of hybrid machine-learning approaches used in email spam filtering and malware detection. The paper examines related studies, commonly used hybrid algorithms, and datasets reported in the literature.
Enhancing Email Security: A Hybrid Machine Learning Approach for Spam and Malware Detection
World Journal of Advanced Engineering Technology and Sciences, 12(1), 187–200
A study of a hybrid Naïve Bayes–Artificial Neural Network approach. The article reports a spam/ham classification experiment and compares the hybrid model with individual classifiers; its stated dataset and task should be described according to the paper’s methodology when citing the findings.
RESEARCH REPORTS & MATERIALS
Related research outputs
Explainable AI (XAI) for Email Spam and Phishing Detection
The MSc Group 08 final report documents collaborative work on explainable email spam and phishing detection. This is a group project; the report should be presented as collaborative work rather than an individual publication.
Read group reportOptimization of BERT, RoBERTa, and DeBERTa with XGBoost for Explainable Email Spam and Phishing Detection: A Comparative Analysis
Compares transformer representations paired with XGBoost for email classification and examines SHAP and LIME for model interpretation. The reported experiments use a controlled subset of public email corpora; performance figures should be read alongside the dataset, split, and evaluation details in the report.
Read comparative paperExperimental Findings Summary Report
A supporting report summarising experimental findings for the transformer–XGBoost comparison. It is presented as a technical report, not as a separate peer-reviewed publication.
Read findings reportLocal PDF links expect the files to be placed in the public/documents directory with the filenames shown in the links. If your project uses different filenames or storage, update these paths before deploying.
RESEARCH PRINCIPLES
Transparent methods. Carefully qualified claims.
Model performance should be interpreted in the context of the dataset, class distribution, train/test split, baseline models, evaluation metrics, and limitations. In particular, high accuracy alone does not establish reliable detection of rare or evolving threats.
I aim to make research outputs discoverable and to distinguish peer-reviewed articles from group-project reports, working papers, and supplementary experimental documents.
Discuss research or collaboration