Waleed M. Ismael
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Waleed M. Ismael has a total of 26 co-authors. Here are the top co-authors by number of shared publications:
Ateeq Ur Rehman
Computer Science and Engineering, Saveetha School of Engineering, Saveetha Institute of Medical and Technical Sciences, Chennai, Tamilnadu, India
4 shared publicationsFahad Ahmed
School of Computer Science, National College of Business Administration and Economics, Lahore, Pakistan
3 shared publications- Muhammad Adnan Khan
Department of Software, Faculty of Artificial Intelligence and Software, Gachon University
3 shared publications Naila Sammar Naz
School of Computer Science, National College of Business Administration and Economics, Lahore, Pakistan
2 shared publicationsAkash Biju
Department of Mathematics, Amrita School of Physical Sciences Coimbatore, Amrita Vishwa Vidyapeetham, Coimbatore, India
1 shared publicationAli Zaman Malik
Department of Computer Science, National College of Business Administration and Economics, Lahore, Pakistan
1 shared publication
Latest Publications
- Open Access
- 3 October 2026
- Published by Springer Nature in Scientific Reports
Deep feature selection and neural architectures for robust darknet traffic classification
Accurately categorizing darknet traffic remains a significant cybersecurity challenge because of the encrypted and anonymous nature of communications made possible by technologies like Virtual Private Networks (VPNs) and The Onion Router (Tor). This work proposes a structured deep learning (DL)-based framework for darknet traffic classification using both individual and combinatorial feature selection (FS) techniques on the CIC-Darknet2020 dataset. We employed five FS techniques, Mutual Information (MI), Random Forest (RF), Recursive Feature Elimination (RFE), Autoencoder-based FS, and XGBoost (XGB), to identify the top 20 discriminative features from the dataset. 31 unique feature subsets with both individual and intersected feature combinations were generated using a combinatorial FS approach. The CNN model trained on RFE-selected features outperformed all other models evaluated, with a testing accuracy of 95.2% and a logarithmic loss of 0.1257. Kuncheva’s Consistency Index was used to measure the stability of each FS technique’s own feature rankings, and the results showed that tree-based methods yielded more repeatable rankings than the Autoencoder-based approach. Using multi-seed retraining and formal statistical significance testing, this ranking was further validated. Furthermore, it was shown that the top-performing configurations were practically suitable for operational deployment because they were computationally efficient and well-calibrated for real-time inference. Additionally, the results demonstrate that optimized FS techniques greatly enhance classification performance when compared to increasing architectural complexity. Although combinatorial FS methods outperformed MI and Autoencoder-based strategies in terms of robustness and generalization, the best-performing models were primarily derived from individual tree-based FS techniques. The analysis also discovered that Tor traffic was the most difficult to classify due to its encrypted and obfuscated communication patterns. The proposed framework offers a solid foundation for cybersecurity and scalable darknet traffic analysis, and it demonstrates a strong capacity for generalization.
- 3 October 2026
- Published by Springer Nature in Scientific Reports
Open AccessDeep feature selection and neural architectures for robust darknet traffic classification
Accurately categorizing darknet traffic remains a significant cybersecurity challenge because of the encrypted and anonymous nature of communications made possible by technologies like Virtual Private Networks (VPNs) and The Onion Router (Tor). This work proposes a structured deep learning (DL)-based framework for darknet traffic classification using both individual and combinatorial feature selection (FS) techniques on the CIC-Darknet2020 dataset. We employed five FS techniques, Mutual Information (MI), Random Forest (RF), Recursive Feature Elimination (RFE), Autoencoder-based FS, and XGBoost (XGB), to identify the top 20 discriminative features from the dataset. 31 unique feature subsets with both individual and intersected feature combinations were generated using a combinatorial FS approach. The CNN model trained on RFE-selected features outperformed all other models evaluated, with a testing accuracy of 95.2% and a logarithmic loss of 0.1257. Kuncheva’s Consistency Index was used to measure the stability of each FS technique’s own feature rankings, and the results showed that tree-based methods yielded more repeatable rankings than the Autoencoder-based approach. Using multi-seed retraining and formal statistical significance testing, this ranking was further validated. Furthermore, it was shown that the top-performing configurations were practically suitable for operational deployment because they were computationally efficient and well-calibrated for real-time inference. Additionally, the results demonstrate that optimized FS techniques greatly enhance classification performance when compared to increasing architectural complexity. Although combinatorial FS methods outperformed MI and Autoencoder-based strategies in terms of robustness and generalization, the best-performing models were primarily derived from individual tree-based FS techniques. The analysis also discovered that Tor traffic was the most difficult to classify due to its encrypted and obfuscated communication patterns. The proposed framework offers a solid foundation for cybersecurity and scalable darknet traffic analysis, and it demonstrates a strong capacity for generalization.
- Open Access
- 10 April 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1)
Honey yield prediction and neonicotinoid risk assessment utilizing a machine learning framework in smart agriculture
Reduction in pollinator abundance (predominantly honeybees) stemming from environmental and chemical stressors notably neonicotinoid pesticides poses serious threats to biodiversity and agricultural productivity. This study presents a scalable machine learning framework to foresee honey yield and assess the impact of neonicotinoid exposure. Drawing upon a curated dataset, of 825 records encompassing 16 agro-environmental and chemical parameters including colony counts, yield metrics, and pesticide residue concentrations. Subsequently, we assessed 13 classification models, marked improvement was evident, as ensemble models consistently outperformed individual learners. Notably, our proposed voting classifier (AgriBuzzEnsemble), which synergistically fuses Support Vector Machine (SVM) and Gaussian Naive Bayes (GNB), surpassed all baseline models, showcasing robust accuracy of 98%, a Matthews Correlation Coefficient (MCC) of 0.9751, a specificity of 0.9917, and an ROC AUC of 0.9985. Data preparation encompassed missing value imputation, outlier detection, feature scaling, and SMOTE based class balancing. For subsequent analysis, we employed Z-score filtering to detect and remove outliers, followed by a log1p transformation to mitigate skewness adhering to standard and well established preprocessing standards. Correlation analysis and statistical tests comprising McNemar's Test, ANOVA, and Tukey's HSD validated model reliability and confirmed the negative association between neonicotinoid burden and honey yield. Our proposed framework AgriBuzzEnsemble facilitates precision focused beekeeping by identifying yield risk zones and can be generalized to other regions facing pollinator stress, offering a robust and interpretable solution for sustainable agricultural planning.
- 10 April 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1)
Open AccessHoney yield prediction and neonicotinoid risk assessment utilizing a machine learning framework in smart agriculture
Reduction in pollinator abundance (predominantly honeybees) stemming from environmental and chemical stressors notably neonicotinoid pesticides poses serious threats to biodiversity and agricultural productivity. This study presents a scalable machine learning framework to foresee honey yield and assess the impact of neonicotinoid exposure. Drawing upon a curated dataset, of 825 records encompassing 16 agro-environmental and chemical parameters including colony counts, yield metrics, and pesticide residue concentrations. Subsequently, we assessed 13 classification models, marked improvement was evident, as ensemble models consistently outperformed individual learners. Notably, our proposed voting classifier (AgriBuzzEnsemble), which synergistically fuses Support Vector Machine (SVM) and Gaussian Naive Bayes (GNB), surpassed all baseline models, showcasing robust accuracy of 98%, a Matthews Correlation Coefficient (MCC) of 0.9751, a specificity of 0.9917, and an ROC AUC of 0.9985. Data preparation encompassed missing value imputation, outlier detection, feature scaling, and SMOTE based class balancing. For subsequent analysis, we employed Z-score filtering to detect and remove outliers, followed by a log1p transformation to mitigate skewness adhering to standard and well established preprocessing standards. Correlation analysis and statistical tests comprising McNemar's Test, ANOVA, and Tukey's HSD validated model reliability and confirmed the negative association between neonicotinoid burden and honey yield. Our proposed framework AgriBuzzEnsemble facilitates precision focused beekeeping by identifying yield risk zones and can be generalized to other regions facing pollinator stress, offering a robust and interpretable solution for sustainable agricultural planning.
- Open Access
- 18 February 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1)
Ensemble machine learning for proactive android ransomware detection using network traffic
Android ransomware has emerged as a major threat to mobile ecosystems, leveraging obfuscated payloads and dynamic command-and-control channels to evade conventional detection systems. Existing approaches often rely on static, batch-trained models that lack adaptability to evolving threat behaviors, resulting in degraded accuracy over time due to concept drift. This presents a critical challenge for real-time deployment, as new ransomware variants continually mutate their signatures and alter network traffic patterns to evade detection. To bridge this gap, this study proposes a robust ensemble-based machine learning framework for proactive detection of Android ransomware using network traffic metadata. The framework integrates advanced classifiers, including Light Gradient Boosting Machine, eXtreme Gradient Boosting Machine, and Random Forest, with Synthetic Minority Oversampling Technique enhanced stratified cross-validation to mitigate class imbalance and improve generalizability. Furthermore, explainable artificial intelligence methods such as SHapley Additive exPlanations and Local Interpretable Model-Agnostic Explanations are employed to enhance interpretability and analyst trust. In the context of ransomware detection, the importance of online learning lies in its ability to adapt to evolving threat patterns in real time. Ransomware frequently mutates payload signatures and obfuscates behavioral traces, causing traditional models to deteriorate under changing data distributions. To address this, we conducted a concept drift evaluation using an incremental LightGBM model, tested on chronologically partitioned traffic data across five temporal blocks. This approach enables continuous adaptation to new data streams without requiring full retraining, thereby maintaining detection robustness and reducing false negatives in production. Experimental results on a balanced dataset demonstrate that LightGBM achieves the highest classification performance, indicating the efficacy and adaptability of the proposed framework for real-time Android ransomware mitigation in dynamic network environments.
- 18 February 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1)
Open AccessEnsemble machine learning for proactive android ransomware detection using network traffic
Android ransomware has emerged as a major threat to mobile ecosystems, leveraging obfuscated payloads and dynamic command-and-control channels to evade conventional detection systems. Existing approaches often rely on static, batch-trained models that lack adaptability to evolving threat behaviors, resulting in degraded accuracy over time due to concept drift. This presents a critical challenge for real-time deployment, as new ransomware variants continually mutate their signatures and alter network traffic patterns to evade detection. To bridge this gap, this study proposes a robust ensemble-based machine learning framework for proactive detection of Android ransomware using network traffic metadata. The framework integrates advanced classifiers, including Light Gradient Boosting Machine, eXtreme Gradient Boosting Machine, and Random Forest, with Synthetic Minority Oversampling Technique enhanced stratified cross-validation to mitigate class imbalance and improve generalizability. Furthermore, explainable artificial intelligence methods such as SHapley Additive exPlanations and Local Interpretable Model-Agnostic Explanations are employed to enhance interpretability and analyst trust. In the context of ransomware detection, the importance of online learning lies in its ability to adapt to evolving threat patterns in real time. Ransomware frequently mutates payload signatures and obfuscates behavioral traces, causing traditional models to deteriorate under changing data distributions. To address this, we conducted a concept drift evaluation using an incremental LightGBM model, tested on chronologically partitioned traffic data across five temporal blocks. This approach enables continuous adaptation to new data streams without requiring full retraining, thereby maintaining detection robustness and reducing false negatives in production. Experimental results on a balanced dataset demonstrate that LightGBM achieves the highest classification performance, indicating the efficacy and adaptability of the proposed framework for real-time Android ransomware mitigation in dynamic network environments.
- Top Cited PapersOpen Access
- 5 January 2026
- Published by Springer Nature in BMC Medical Imaging
- Vol. 26 (1)
Explainable artificial intelligence (XAI) in medical imaging: a systematic review of techniques, applications, and challenges
Explainable Artificial Intelligence (XAI) is crucial for enhancing transparency and trustworthiness, as well as for developing AI-based diagnostic systems
- 5 January 2026
- Published by Springer Nature in BMC Medical Imaging
- Vol. 26 (1)
Top Cited PapersOpen AccessExplainable artificial intelligence (XAI) in medical imaging: a systematic review of techniques, applications, and challenges
Explainable Artificial Intelligence (XAI) is crucial for enhancing transparency and trustworthiness, as well as for developing AI-based diagnostic systems
- Open Access
- 26 November 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 42118
Enhancing smart city mobility through real time explainable AI in autonomous vehicles
The quick advancement of Autonomous Vehicular Networks (AVNs) shows remarkable potential to transform urban transportation systems in smart cities. This transformational process faces several crucial issues, primarily due to the lack of clear decision-making, concerns about public confidence, and the need for prompt protective measures in various contexts. The implementation of AVNs depends on resolving current adversities, as these difficulties affect both system safety, user trust, and performance reliability. Traditional AVN development focused on enhancing technical capabilities, such as reliability, but failed to adequately address issues with decision transparency and interpretability. The current systems fall short because they lack an understanding of how Autonomous Vehicles (AVs) generate decisions in real-time urban conditions, which impedes public confidence and broader adoption. To address these limitations, this study integrates You Only Look Once, V5 (YOLOv5), a fast and lightweight object detection model well-suited for AVs, alongside Explainable AI (XAI) techniques to ensure interpretability and transparency. In this research, an XAI-based YOLOv5 model is proposed to enable real-time, explainable decision-making. Its objectives are to increase transparency, increase safety, and gain public acceptance for connecting the AVNs to smart city systems. The proposed model achieves an accuracy of 99% with a miss rate of 1%, thereby enhancing classification accuracy and public confidence. The proposed work also aims to foster public trust in AVNs within smart city ecosystems by making AI decisions more transparent and interpretable.
- 26 November 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 42118
Open AccessEnhancing smart city mobility through real time explainable AI in autonomous vehicles
The quick advancement of Autonomous Vehicular Networks (AVNs) shows remarkable potential to transform urban transportation systems in smart cities. This transformational process faces several crucial issues, primarily due to the lack of clear decision-making, concerns about public confidence, and the need for prompt protective measures in various contexts. The implementation of AVNs depends on resolving current adversities, as these difficulties affect both system safety, user trust, and performance reliability. Traditional AVN development focused on enhancing technical capabilities, such as reliability, but failed to adequately address issues with decision transparency and interpretability. The current systems fall short because they lack an understanding of how Autonomous Vehicles (AVs) generate decisions in real-time urban conditions, which impedes public confidence and broader adoption. To address these limitations, this study integrates You Only Look Once, V5 (YOLOv5), a fast and lightweight object detection model well-suited for AVs, alongside Explainable AI (XAI) techniques to ensure interpretability and transparency. In this research, an XAI-based YOLOv5 model is proposed to enable real-time, explainable decision-making. Its objectives are to increase transparency, increase safety, and gain public acceptance for connecting the AVNs to smart city systems. The proposed model achieves an accuracy of 99% with a miss rate of 1%, thereby enhancing classification accuracy and public confidence. The proposed work also aims to foster public trust in AVNs within smart city ecosystems by making AI decisions more transparent and interpretable.
- Open Access
- 15 October 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 1-27
Privacy preserving skin cancer diagnosis through federated deep learning and explainable AI
The classification of human skin disorders, particularly benign and malignant skin cancer, is thoroughly examined in this study with a focus on protecting data privacy. Traditional visual diagnosis of skin disorders is often subjective and complicated by the varying colors, textures, and shapes of lesions. To address these challenges, we propose a privacy-preserving and explainable deep learning (DL) architecture that leverages secure federated learning (FL) on distributed medical data sources without exposing private patient information, ensuring compliance with data protection regulations. Real-world decentralized scenarios are simulated by dividing a skin image dataset into two classes and distributing it among three clients. The Federated Averaging (FedAvg) method is employed to train the VGG19 model—a well-established convolutional neural network (CNN)—over 25 federated communication rounds, after pretraining on ImageNet and fine-tuning for binary classification. To enhance robustness and diversity, dermatology datasets, such as Kaggle, are often used in similar studies for performance evaluation. Additionally, explainable AI (XAI) techniques, such as Gradient-weighted Class Activation Mapping (Grad-CAM), are incorporated to improve transparency and assist clinicians in visualizing and interpreting the model’s decision-making process. Experimental results demonstrate that the federated approach maintains data privacy while achieving high classification performance. This work highlights the potential of combining explainability and FL to develop reliable and privacy-conscious AI solutions for dermatological diagnosis.
- 15 October 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 1-27
Open AccessPrivacy preserving skin cancer diagnosis through federated deep learning and explainable AI
The classification of human skin disorders, particularly benign and malignant skin cancer, is thoroughly examined in this study with a focus on protecting data privacy. Traditional visual diagnosis of skin disorders is often subjective and complicated by the varying colors, textures, and shapes of lesions. To address these challenges, we propose a privacy-preserving and explainable deep learning (DL) architecture that leverages secure federated learning (FL) on distributed medical data sources without exposing private patient information, ensuring compliance with data protection regulations. Real-world decentralized scenarios are simulated by dividing a skin image dataset into two classes and distributing it among three clients. The Federated Averaging (FedAvg) method is employed to train the VGG19 model—a well-established convolutional neural network (CNN)—over 25 federated communication rounds, after pretraining on ImageNet and fine-tuning for binary classification. To enhance robustness and diversity, dermatology datasets, such as Kaggle, are often used in similar studies for performance evaluation. Additionally, explainable AI (XAI) techniques, such as Gradient-weighted Class Activation Mapping (Grad-CAM), are incorporated to improve transparency and assist clinicians in visualizing and interpreting the model’s decision-making process. Experimental results demonstrate that the federated approach maintains data privacy while achieving high classification performance. This work highlights the potential of combining explainability and FL to develop reliable and privacy-conscious AI solutions for dermatological diagnosis.