James Ben Hayfron-Acquah
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
James Ben Hayfron-Acquah has a total of 14 co-authors. Here are the top co-authors by number of shared publications:
- Frimpong Twum
Department of Computer Science, Kwame Nkrumah University of Science and Technology, Kumasi, Ghana
2 shared publications - James Ben Hayfron-Acquah
Faculty of Physical and Computational Sciences, Kwame Nkrumah University of Science and Technology (KNUST)
2 shared publications - Richard Apau
Kwame Nkrumah University of Science and Technology, Ghana
2 shared publications Abdul-Basit Danjoe Munkaila
Department of Logistics and Procurement Management, Faculty of Business, Tamale Technical University
1 shared publicationBoakye Agyemang
Koforidua Technical University, Faculty of Applied Sciences and Technology, Applied Mathematics Department
1 shared publicationEdwin Mends-Brew
Department of Applied Mathematics and Statistics, Accra Technical University
1 shared publication
Latest Publications
- Open Access
- 20 August 2026
- Published by Elsevier in SSRN Electronic Journal
SC-SecMHAD: Supervised Contrastive Learning with SecBERT for Fine-Grained Multi-Class HTTP Attack Detection
The rapid expansion of web applications and Application Programming Interfaces (APIs) has increased the application layer attack surface. This has created a pressing demand for detection systems capable of distinguishing between multiple HyperText Transfer Protocol (HTTP) attack categories. Although deep learning has significantly advanced web attack detection, most existing studies frame the detection problem as a binary task by merely separating malicious traffic from benign traffic. Additionally, there is limited evidence on whether modern contrastive learning objectives can improve the geometric separation of attack classes in the embedding space. This paper proposes SC-SecMHAD for fine-grained HTTP attack detection. The proposed framework integrates a SecBERT encoder with a three-layer projection head optimised under a supervised contrastive objective and linear classifier head into a single end-to-end pipeline. Results show that our proposed framework achieves an accuracy of 0.9892, a macro-F1 score of 0.9766, and a macro AUROC of 0.9999 on the SR-BH 2020 dataset. These results carry direct operational value. Unlike binary systems that only flag HTTP requests as malicious or benign, SC-SecMHAD classifies which of eight specific attack families an attack belongs to, giving security teams the granular information needed to trigger attack-specific containment policies. A Silhouette Score and Davies-Bouldin Index of 0.9116 and 0.149, respectively, measured in the 128-dimensional projection space, provide quantitative confirmation that the learned embedding space exhibits well-separated per-class clusters. The proposed framework achieves a mean inference latency of 18.24 milliseconds per request in an offline batch configuration.
- 20 August 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessSC-SecMHAD: Supervised Contrastive Learning with SecBERT for Fine-Grained Multi-Class HTTP Attack Detection
The rapid expansion of web applications and Application Programming Interfaces (APIs) has increased the application layer attack surface. This has created a pressing demand for detection systems capable of distinguishing between multiple HyperText Transfer Protocol (HTTP) attack categories. Although deep learning has significantly advanced web attack detection, most existing studies frame the detection problem as a binary task by merely separating malicious traffic from benign traffic. Additionally, there is limited evidence on whether modern contrastive learning objectives can improve the geometric separation of attack classes in the embedding space. This paper proposes SC-SecMHAD for fine-grained HTTP attack detection. The proposed framework integrates a SecBERT encoder with a three-layer projection head optimised under a supervised contrastive objective and linear classifier head into a single end-to-end pipeline. Results show that our proposed framework achieves an accuracy of 0.9892, a macro-F1 score of 0.9766, and a macro AUROC of 0.9999 on the SR-BH 2020 dataset. These results carry direct operational value. Unlike binary systems that only flag HTTP requests as malicious or benign, SC-SecMHAD classifies which of eight specific attack families an attack belongs to, giving security teams the granular information needed to trigger attack-specific containment policies. A Silhouette Score and Davies-Bouldin Index of 0.9116 and 0.149, respectively, measured in the 128-dimensional projection space, provide quantitative confirmation that the learned embedding space exhibits well-separated per-class clusters. The proposed framework achieves a mean inference latency of 18.24 milliseconds per request in an offline batch configuration.
- Open Access
- 12 January 2026
- Published by Science Publishing Corporation in International Journal of Advanced Statistics and Probability
- Vol. 13 (1) , 1-8
Predictive Modeling of Ghana’s Economic Growth: a Comparative Analysis of Support Vector Regressionand Ordinary Least Squares Regression UsingForeign Direct Investment Data
The contribution of Foreign Direct Investment (FDI) to economic growth remains a central issue in development economics, particularly for emerging economies that are undergoing macroeconomic adjustment. This study examines the relationship between FDI and Ghana’s Gross Domestic Product (GDP) growth by conducting a comparative assessment of Ordinary Least Squares Regression (OLSR) and Support Vector Regression (SVR). Annual data spanning 1996–2023 were obtained from the World Bank and OECD national accounts databases. Model performance was evaluated using out-of-sample Root Mean Square Error (RMSE) and coefficient of determination (R²) to distinguish predictive accuracy from explanatory power. The results indicate that FDI exerts a statistically significant and positive effect on GDP growth under the OLSR framework, with an R² value of 0.47, suggesting moderate explanatory strength. In contrast, the SVR model , implemented with a radial basis function kernel and tuned via cross-validation, achieved a marginally lower prediction error (RMSE = 0.754) , but lower explanatory power (R² = 0.32). These findings highlight a clear trade-off between predictive accuracy and interpretability. The study emphasizes that the single-predictor specification serves a methodological purpose rather than a comprehensive economic representation, and that results should be interpreted within this constrained framework. Generally, the analysis shows the complementary roles of machine learning and econometric approaches in applied economic modeling and provides evidence-based guidance for model selection in macroeconomic forecasting and policy analysis in Ghana.
- 12 January 2026
- Published by Science Publishing Corporation in International Journal of Advanced Statistics and Probability
- Vol. 13 (1) , 1-8
Open AccessPredictive Modeling of Ghana’s Economic Growth: a Comparative Analysis of Support Vector Regressionand Ordinary Least Squares Regression UsingForeign Direct Investment Data
The contribution of Foreign Direct Investment (FDI) to economic growth remains a central issue in development economics, particularly for emerging economies that are undergoing macroeconomic adjustment. This study examines the relationship between FDI and Ghana’s Gross Domestic Product (GDP) growth by conducting a comparative assessment of Ordinary Least Squares Regression (OLSR) and Support Vector Regression (SVR). Annual data spanning 1996–2023 were obtained from the World Bank and OECD national accounts databases. Model performance was evaluated using out-of-sample Root Mean Square Error (RMSE) and coefficient of determination (R²) to distinguish predictive accuracy from explanatory power. The results indicate that FDI exerts a statistically significant and positive effect on GDP growth under the OLSR framework, with an R² value of 0.47, suggesting moderate explanatory strength. In contrast, the SVR model , implemented with a radial basis function kernel and tuned via cross-validation, achieved a marginally lower prediction error (RMSE = 0.754) , but lower explanatory power (R² = 0.32). These findings highlight a clear trade-off between predictive accuracy and interpretability. The study emphasizes that the single-predictor specification serves a methodological purpose rather than a comprehensive economic representation, and that results should be interpreted within this constrained framework. Generally, the analysis shows the complementary roles of machine learning and econometric approaches in applied economic modeling and provides evidence-based guidance for model selection in macroeconomic forecasting and policy analysis in Ghana.
- 18 December 2024
- Published by Springer Nature
- p. 307-325
A Secure and High Embedding Capacity Image Steganography Scheme Using Image Enhancement Technique, Compression, and Genetic Algorithm
- 18 December 2024
- Published by Springer Nature
- p. 307-325
A Secure and High Embedding Capacity Image Steganography Scheme Using Image Enhancement Technique, Compression, and Genetic Algorithm