Mohamed Abdulnabi
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Mohamed Abdulnabi has a total of 44 co-authors. Here are the top co-authors by number of shared publications:
- Aitizaz Ali
Asia Pacific University of Technology and Innovation
4 shared publications Aitizaz Ali
School of Technology, Asia Pacific University of Technology & Innovation (APU), Kuala Lumpur, Malaysia
3 shared publications- Abdullah Mohammed Alqahtani
College of Engineering and Computer Science, Jazan University, Jazan 45142, Saudi Arabia
2 shared publications - Arshad Ali
Islamic University of Madinah
2 shared publications - Hamayun Khan
Department of Computer Science, Faculty of Computer Science & IT, Superior University, Lahore, Pakistan
2 shared publications - Irfanud Din
Department of Computer Science, New Uzbekistan University, Tashkent, Uzbekistan
2 shared publications
Latest Publications
- Open Access
- 15 October 2026
- Published by Springer Nature in Discover Artificial Intelligence
- Vol. 6  (1) , 1502
A modified dual-task DenseNet121 with Grad-CAM interpretability for brain tumor classification from MRI
Diagnosing brain tumors from MRI scans is technically challenging but necessary. Early and correct characterization provides direct indications for subsequent treatment of and survival of the patient. While deep learning models show accuracy, most of them focus on just one classification task lacking practical tools for clinical use. We introduce Modified DenseNet121 (Dual-Task framework built on a modified DenseNet121 with a custom regularization stack. This is important to detect abnormalities (Normal vs. Abnormal) and identify three tumor types (Glioma, Meningioma, and Pituitary) in a single architecture. The model incorporates Gradient-weighted Class Activation Mapping (Grad-CAM) for qualitative spatial interpretability, and the model is set up within a PyQt5 desktop research prototype illustrating real-time dual prediction with typical inference latency of 80–130 ms (mean: 185.80 ± 402.83 ms; median: 127.99 ms; 95th percentile: 134.65 ms), probability visualization, Grad-CAM heatmap overlay, and feedback recording from the physician. Elaborate benchmarking experiments were performed against six competing CNN architectures DenseNet121 (baseline), ConvNeXtSmall, ConvNeXtTiny, EfficientNetB0, NASNetMobile and Xception over two large, publicly available MRI datasets: dataset 1, Kaggle, holding 7023 images; and dataset 2, Mendeley, having 12,064 images. Modified DenseNet121 achieved 99.85% binary accuracy and 99.00% multiclass accuracy on dataset 1, and 99.62% binary accuracy and 99.00% multiclass accuracy on dataset 2, achieving competitive performance with all benchmark architectures across both datasets. The zero-shot evaluation on an additional publicly available external MRI dataset achieved 99.75% binary accuracy and 99.20% multiclass accuracy, demonstrating transfer performance under an external evaluation setting. However, because dedicated image-level overlap analysis was not performed, these results are interpreted as evidence of external transfer performance rather than definitive evidence of complete dataset independence. The experimental results demonstrate high predictive performance across two large publicly available datasets and verify the technical feasibility of Modified DenseNet121 as a robust research prototype. This study offers reproducible, interpretable, and research prototype foundation for future clinical translation of neuro-oncological computer-aided diagnosis systems.
- 15 October 2026
- Published by Springer Nature in Discover Artificial Intelligence
- Vol. 6  (1) , 1502
Open AccessA modified dual-task DenseNet121 with Grad-CAM interpretability for brain tumor classification from MRI
Diagnosing brain tumors from MRI scans is technically challenging but necessary. Early and correct characterization provides direct indications for subsequent treatment of and survival of the patient. While deep learning models show accuracy, most of them focus on just one classification task lacking practical tools for clinical use. We introduce Modified DenseNet121 (Dual-Task framework built on a modified DenseNet121 with a custom regularization stack. This is important to detect abnormalities (Normal vs. Abnormal) and identify three tumor types (Glioma, Meningioma, and Pituitary) in a single architecture. The model incorporates Gradient-weighted Class Activation Mapping (Grad-CAM) for qualitative spatial interpretability, and the model is set up within a PyQt5 desktop research prototype illustrating real-time dual prediction with typical inference latency of 80–130 ms (mean: 185.80 ± 402.83 ms; median: 127.99 ms; 95th percentile: 134.65 ms), probability visualization, Grad-CAM heatmap overlay, and feedback recording from the physician. Elaborate benchmarking experiments were performed against six competing CNN architectures DenseNet121 (baseline), ConvNeXtSmall, ConvNeXtTiny, EfficientNetB0, NASNetMobile and Xception over two large, publicly available MRI datasets: dataset 1, Kaggle, holding 7023 images; and dataset 2, Mendeley, having 12,064 images. Modified DenseNet121 achieved 99.85% binary accuracy and 99.00% multiclass accuracy on dataset 1, and 99.62% binary accuracy and 99.00% multiclass accuracy on dataset 2, achieving competitive performance with all benchmark architectures across both datasets. The zero-shot evaluation on an additional publicly available external MRI dataset achieved 99.75% binary accuracy and 99.20% multiclass accuracy, demonstrating transfer performance under an external evaluation setting. However, because dedicated image-level overlap analysis was not performed, these results are interpreted as evidence of external transfer performance rather than definitive evidence of complete dataset independence. The experimental results demonstrate high predictive performance across two large publicly available datasets and verify the technical feasibility of Modified DenseNet121 as a robust research prototype. This study offers reproducible, interpretable, and research prototype foundation for future clinical translation of neuro-oncological computer-aided diagnosis systems.
- Open Access
- 10 March 2026
- Published by Springer Nature in Discover Computing
- Vol. 29  (1) , 147
A systematic literature review on the accuracy of machine learning based DDoS detection techniques in Software Defined Networking
Software Defined Networking (SDN) enhances programmability and scalability but is significantly susceptible to Distributed Denial of Service (DDoS) attacks. This research methodically assesses machine learning (ML) detection strategies for DDoS attacks in SDN, emphasizing detection efficacy, datasets, and evaluation metrics. Employing the PRISMA framework, 36 main papers published from 2018 to 2023 were discovered and examined. The results indicate that Random Forest (RF), Support Vector Machine (SVM), Decision Trees (DT), and hybrid deep learning methodologies regularly attain high accuracy, often exceeding 99%, in identifying DDoS traffic. Nevertheless, the majority of research provide findings only based on accuracy, with little consideration of variability, statistical significance, and real-world applicability. A bibliometric analysis of worldwide contributions is presented to offer research context, emphasizing India, China, and Saudi Arabia as prominent contributors. This research underscores the need for multi-factor assessment (precision, recall, F1-score), rigorous validation across varied datasets, and real-time scalability assessment. The research emphasizes the significance of feature selection and extraction in enhancing performance, the prevalence of control-plane implementations, and the dependence on comparison datasets like CICDDoS2019. Challenges persist due to limitations like dataset imbalance, insufficient real-world testing, and inadequate reporting of variability measurements. Future initiatives include the development of lightweight and adaptable models, the exploration of transfer and adversarial learning, and the validation of methodologies in extensive SDN-IoT networks. These results elucidate the strengths and weaknesses of existing ML-based detection methods in SDN and delineate prospective avenues for future study.
- 10 March 2026
- Published by Springer Nature in Discover Computing
- Vol. 29  (1) , 147
Open AccessA systematic literature review on the accuracy of machine learning based DDoS detection techniques in Software Defined Networking
Software Defined Networking (SDN) enhances programmability and scalability but is significantly susceptible to Distributed Denial of Service (DDoS) attacks. This research methodically assesses machine learning (ML) detection strategies for DDoS attacks in SDN, emphasizing detection efficacy, datasets, and evaluation metrics. Employing the PRISMA framework, 36 main papers published from 2018 to 2023 were discovered and examined. The results indicate that Random Forest (RF), Support Vector Machine (SVM), Decision Trees (DT), and hybrid deep learning methodologies regularly attain high accuracy, often exceeding 99%, in identifying DDoS traffic. Nevertheless, the majority of research provide findings only based on accuracy, with little consideration of variability, statistical significance, and real-world applicability. A bibliometric analysis of worldwide contributions is presented to offer research context, emphasizing India, China, and Saudi Arabia as prominent contributors. This research underscores the need for multi-factor assessment (precision, recall, F1-score), rigorous validation across varied datasets, and real-time scalability assessment. The research emphasizes the significance of feature selection and extraction in enhancing performance, the prevalence of control-plane implementations, and the dependence on comparison datasets like CICDDoS2019. Challenges persist due to limitations like dataset imbalance, insufficient real-world testing, and inadequate reporting of variability measurements. Future initiatives include the development of lightweight and adaptable models, the exploration of transfer and adversarial learning, and the validation of methodologies in extensive SDN-IoT networks. These results elucidate the strengths and weaknesses of existing ML-based detection methods in SDN and delineate prospective avenues for future study.
- 2 January 2026
- Published by Springer Nature
- p. 519-531
Containerizing Penetration Testing Tools Using Docker Containers
- 2 January 2026
- Published by Springer Nature
- p. 519-531
Containerizing Penetration Testing Tools Using Docker Containers
- Open Access
- 29 December 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15  (1) , 45035
Intelligent delignification: leveraging explainable AI for ozone transport modeling and optimization
Biomass is mainly composed of cellulose, hemicellulose, and lignin, where lignin is almost one-third of the amount of biomass. Lignin is removed from the biomass matrix because its complex, recalcitrant structure acts as a physical and chemical barrier, impeding the accessibility of reagents and enzymes to cellulose and hemicellulose. Lignin sterically hinders hydrolysis and can also non-productively bind enzymes, thereby reducing overall efficiency in biofuel production. Several techniques are used to remove lignin from the biomass matrix, where ozonation is the most novel, emerging, and ecofriendly technique. This study investigates the application of machine learning (ML) techniques to predict lignin removal efficiency during ozonation pretreatment an emerging and eco-friendly delignification method. Experimental data from ozonation-based lignin removal were used to train and evaluate 19 regression models using the PyCaret framework. Among these, the Extra Trees Regressor demonstrated the highest predictive accuracy for delignification outcomes. Feature importance was further interpreted using SHapley Additive exPlanations (SHAP) to quantify the contribution of each process variable. The results reveal that ML models can effectively capture the complex relationships governing ozonation-based delignification, offering valuable insights into optimizing operational parameters. This work highlights the potential of ML as a predictive and interpretative tool in chemical engineering, paving the way for more efficient, data-driven approaches to biomass valorization and sustainable biofuel.
- 29 December 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15  (1) , 45035
Open AccessIntelligent delignification: leveraging explainable AI for ozone transport modeling and optimization
Biomass is mainly composed of cellulose, hemicellulose, and lignin, where lignin is almost one-third of the amount of biomass. Lignin is removed from the biomass matrix because its complex, recalcitrant structure acts as a physical and chemical barrier, impeding the accessibility of reagents and enzymes to cellulose and hemicellulose. Lignin sterically hinders hydrolysis and can also non-productively bind enzymes, thereby reducing overall efficiency in biofuel production. Several techniques are used to remove lignin from the biomass matrix, where ozonation is the most novel, emerging, and ecofriendly technique. This study investigates the application of machine learning (ML) techniques to predict lignin removal efficiency during ozonation pretreatment an emerging and eco-friendly delignification method. Experimental data from ozonation-based lignin removal were used to train and evaluate 19 regression models using the PyCaret framework. Among these, the Extra Trees Regressor demonstrated the highest predictive accuracy for delignification outcomes. Feature importance was further interpreted using SHapley Additive exPlanations (SHAP) to quantify the contribution of each process variable. The results reveal that ML models can effectively capture the complex relationships governing ozonation-based delignification, offering valuable insights into optimizing operational parameters. This work highlights the potential of ML as a predictive and interpretative tool in chemical engineering, paving the way for more efficient, data-driven approaches to biomass valorization and sustainable biofuel.
- Open Access
- 8 December 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15  (1) , 43310
Classifying human vs. AI text with machine learning and explainable transformer models
The rapid proliferation of AI-generated text from models such as ChatGPT-3.5 and ChatGPT-4 has raised critical challenges in verifying content authenticity and ensuring ethical use of language technologies. This study presents a comprehensive framework for distinguishing between human-written and GPT-generated text using a combination of machine learning, sequential deep learning, and transformer-based models. A balanced dataset of 20,000 samples was compiled, incorporating diverse linguistic and topical sources. Traditional algorithms and sequential architectures (LSTM, GRU, BiLSTM, BiGRU) were compared against advanced transformer models, including BERT, DistilBERT, ALBERT, and RoBERTa. Experimental findings revealed that RoBERTa achieved the highest performance (Accuracy = 96.1%), outperforming all baselines. Post-hoc temperature scaling (T = 1.476) improved calibration, while threshold tuning (t = 0.957) enhanced precision for high-stakes applications. McNemar’s test with Holm correction confirmed the statistical significance ( p < 0.05) of RoBERTa’s superiority. Efficiency analysis showed optimal trade-offs between accuracy and latency, and 20% pruning demonstrated sustainability potential. Furthermore, LIME and SHAP explainability analyses highlighted linguistic distinctions between AI-generated and human-authored text, and fine-grained error evaluation confirmed model robustness across text lengths. In conclusion, RoBERTa emerges as a reliable, interpretable, and computationally efficient model for detecting AI-generated content.
- 8 December 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15  (1) , 43310
Open AccessClassifying human vs. AI text with machine learning and explainable transformer models
The rapid proliferation of AI-generated text from models such as ChatGPT-3.5 and ChatGPT-4 has raised critical challenges in verifying content authenticity and ensuring ethical use of language technologies. This study presents a comprehensive framework for distinguishing between human-written and GPT-generated text using a combination of machine learning, sequential deep learning, and transformer-based models. A balanced dataset of 20,000 samples was compiled, incorporating diverse linguistic and topical sources. Traditional algorithms and sequential architectures (LSTM, GRU, BiLSTM, BiGRU) were compared against advanced transformer models, including BERT, DistilBERT, ALBERT, and RoBERTa. Experimental findings revealed that RoBERTa achieved the highest performance (Accuracy = 96.1%), outperforming all baselines. Post-hoc temperature scaling (T = 1.476) improved calibration, while threshold tuning (t = 0.957) enhanced precision for high-stakes applications. McNemar’s test with Holm correction confirmed the statistical significance ( p < 0.05) of RoBERTa’s superiority. Efficiency analysis showed optimal trade-offs between accuracy and latency, and 20% pruning demonstrated sustainability potential. Furthermore, LIME and SHAP explainability analyses highlighted linguistic distinctions between AI-generated and human-authored text, and fine-grained error evaluation confirmed model robustness across text lengths. In conclusion, RoBERTa emerges as a reliable, interpretable, and computationally efficient model for detecting AI-generated content.
- 17 October 2025
- Published by Springer Nature
- p. 611-624
Understanding Cognitive Biases in Security Decision-Making and Proposing Effective Mitigation Strategies
- 17 October 2025
- Published by Springer Nature
- p. 611-624
Understanding Cognitive Biases in Security Decision-Making and Proposing Effective Mitigation Strategies
- Open Access
- 31 July 2025
- Published by International Information and Engineering Technology Association in International Journal of Design & Nature and Ecodynamics
- Vol. 20  (7) , 1435-1447
Magnetically Recoverable Fe3O4@BNPs@ZnO-ZnS Nanocomposite with Machine Learning Optimization for Enhanced Photocatalytic Water Purification
- 31 July 2025
- Published by International Information and Engineering Technology Association in International Journal of Design & Nature and Ecodynamics
- Vol. 20  (7) , 1435-1447
Open AccessMagnetically Recoverable Fe3O4@BNPs@ZnO-ZnS Nanocomposite with Machine Learning Optimization for Enhanced Photocatalytic Water Purification
- Open Access
- 24 July 2025
- Published by Springer Nature in Research Square
Intelligent Delignification: Leveraging Explainable AIfor Ozone Transport Modeling and Optimization
A biomass is mainly composed of cellulose, hemicellulose and lignin where lignin is almost one third of the amount of biomass. Lignin is removed from the materials matrix because its presence does not allow reagents to react with cellulose and hemicellulose for use in any biofuel industry. Several techniques are used to remove lignin from the biomass matrix where ozonation is the most novel, emerging and ecofriendly technique. This study presents the feasibility of employing machine learning models in the initial stages of ozonation-based lignin removal from lignocellulosic matrices. Another step was to train and evaluate several machine learning models using the experimental data. Pycaret was used to compare 19 different machine learning models. The extra trees regressor successfully forecasted the delignification. The SHapley Additive eXplanation is used to compute the contribution of features. The results provide light on the extraction process and might pave the way for creating more effective and selective ozonation-based methods of lignin removal, this has important consequences for the development of biofuels, chemicals, and materials. The results of this work should inspire more researchers into the use of machine learning for delignification. Predictive models may be more easily developed with the help of machine learning when used in chemical engineering. To bridge the gap in delignification experimental data and result prediction using artificial intelligence. Machine learning predictions of complicated physical processes are crucial to chemical engineering applications. This approach can be modified for use in future studies to address more delignification problems in chemical engineering.
- 24 July 2025
- Published by Springer Nature in Research Square
Open AccessIntelligent Delignification: Leveraging Explainable AIfor Ozone Transport Modeling and Optimization
A biomass is mainly composed of cellulose, hemicellulose and lignin where lignin is almost one third of the amount of biomass. Lignin is removed from the materials matrix because its presence does not allow reagents to react with cellulose and hemicellulose for use in any biofuel industry. Several techniques are used to remove lignin from the biomass matrix where ozonation is the most novel, emerging and ecofriendly technique. This study presents the feasibility of employing machine learning models in the initial stages of ozonation-based lignin removal from lignocellulosic matrices. Another step was to train and evaluate several machine learning models using the experimental data. Pycaret was used to compare 19 different machine learning models. The extra trees regressor successfully forecasted the delignification. The SHapley Additive eXplanation is used to compute the contribution of features. The results provide light on the extraction process and might pave the way for creating more effective and selective ozonation-based methods of lignin removal, this has important consequences for the development of biofuels, chemicals, and materials. The results of this work should inspire more researchers into the use of machine learning for delignification. Predictive models may be more easily developed with the help of machine learning when used in chemical engineering. To bridge the gap in delignification experimental data and result prediction using artificial intelligence. Machine learning predictions of complicated physical processes are crucial to chemical engineering applications. This approach can be modified for use in future studies to address more delignification problems in chemical engineering.
- Open Access
- 3 April 2025
- Published by Engineering, Technology & Applied Science Research in Engineering, Technology & Applied Science Research
- Vol. 15  (2) , 21279-21283
Forecasting Multi-Level Deep Learning Autoencoder Architecture (MDLAA) for Parametric Prediction based on Convolutional Neural Networks
This study presents a data-driven framework for anomaly detection, which is a significant process in modern computing, as the detection of an abnormal signal can prevent a high-risk decision. The proposed Multi-Level Deep Learning Autoencoder Architecture (MDLAA) is used to encode high dimensional input data using CNNs for anomaly detection in High Dimensional Input Datasets (HDDs). MDLAA is based on unsupervised learning, which has a strong theoretical foundation and is widely used for the detection of anomalies in HDDs, but a few limitations significantly reduce its performance. The proposed MDLAA combines multilevel convolutional layers and data preprocessing. The performance of the proposed model was evaluated on a benchmark dataset. Using feature engineering, the proposed algorithm assists in the detection of anomalies that are present in data structures, especially when compared to the ResNet101 feature extractor. The results show that given adequate data, the proposed technique outperformed other previously implemented deep learning approaches and classification models, showing an overall improvement of 2.3% in terms of MSE, F1-score, precision, and accuracy.
- 3 April 2025
- Published by Engineering, Technology & Applied Science Research in Engineering, Technology & Applied Science Research
- Vol. 15  (2) , 21279-21283
Open AccessForecasting Multi-Level Deep Learning Autoencoder Architecture (MDLAA) for Parametric Prediction based on Convolutional Neural Networks
This study presents a data-driven framework for anomaly detection, which is a significant process in modern computing, as the detection of an abnormal signal can prevent a high-risk decision. The proposed Multi-Level Deep Learning Autoencoder Architecture (MDLAA) is used to encode high dimensional input data using CNNs for anomaly detection in High Dimensional Input Datasets (HDDs). MDLAA is based on unsupervised learning, which has a strong theoretical foundation and is widely used for the detection of anomalies in HDDs, but a few limitations significantly reduce its performance. The proposed MDLAA combines multilevel convolutional layers and data preprocessing. The performance of the proposed model was evaluated on a benchmark dataset. Using feature engineering, the proposed algorithm assists in the detection of anomalies that are present in data structures, especially when compared to the ResNet101 feature extractor. The results show that given adequate data, the proposed technique outperformed other previously implemented deep learning approaches and classification models, showing an overall improvement of 2.3% in terms of MSE, F1-score, precision, and accuracy.
- Open Access
- 2 February 2025
- Published by Engineering, Technology & Applied Science Research in Engineering, Technology & Applied Science Research
- Vol. 15  (1) , 19776-19781
An Enhanced Random Forest (ERF)-based Machine Learning Framework for Resampling, Prediction, and Classification of Mobile Applications using Textual Features
The amount of mobile applications is increasing rapidly, and it is difficult for software developers to identify the numerous key factors that affect their rating and performance. This study presents a machine-learning framework to improve decisions in adding new features to mobile applications and enhancing overall performance. A dataset of app attributes from the Apple AppStore was used, exploiting NLP techniques to preprocess the textual information and develop an Enhanced Random Forest (ERF) framework to assess and forecast ratings for multifunctional apps and investigate the connections between features and user ratings. The ERF model was compared with other renowned ML methods including Decision Trees (DT), Naive Bayes (NB), CNN, and ANN. The experimental results showed that the proposed model predicts app ratings more effectively compared to other complex models. The proposed model achieved precision, recall, and F1-score of 92.76%, 99.33%, and 95.93%, respectively.
- 2 February 2025
- Published by Engineering, Technology & Applied Science Research in Engineering, Technology & Applied Science Research
- Vol. 15  (1) , 19776-19781
Open AccessAn Enhanced Random Forest (ERF)-based Machine Learning Framework for Resampling, Prediction, and Classification of Mobile Applications using Textual Features
The amount of mobile applications is increasing rapidly, and it is difficult for software developers to identify the numerous key factors that affect their rating and performance. This study presents a machine-learning framework to improve decisions in adding new features to mobile applications and enhancing overall performance. A dataset of app attributes from the Apple AppStore was used, exploiting NLP techniques to preprocess the textual information and develop an Enhanced Random Forest (ERF) framework to assess and forecast ratings for multifunctional apps and investigate the connections between features and user ratings. The ERF model was compared with other renowned ML methods including Decision Trees (DT), Naive Bayes (NB), CNN, and ANN. The experimental results showed that the proposed model predicts app ratings more effectively compared to other complex models. The proposed model achieved precision, recall, and F1-score of 92.76%, 99.33%, and 95.93%, respectively.