Ankur Pandey
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Ankur Pandey has a total of 52 co-authors. Here are the top co-authors by number of shared publications:
- Piyush Kumar Shukla
Professor, Department of Computer Science and Engineering & Deputy Dean Research, Amity School of Engineering and Technology (ASET), Amity University, Mumbai, India
5 shared publications Prashant Kumar Shukla
Department of Computer Science and Engineering & Deputy Dean Research (ASET), Amity School of Engineering and Technology (ASET), Amity University, Mumbai, Maharashtra, 410206, India
3 shared publicationsChiranjib Parida
School of Computer Engineering, KIIT-Deemed to be University, Bhubaneswar, 751024, Odisha, India
2 shared publicationsPiyush Kumar Shukla
Department of Computer Science and Engineering, University Institute of Technology, Rajiv Gandhi Proudyogiki Vishwavidyalaya (State Technological University of Madhya Pradesh), Bhopal, Madhya Pradesh, India
2 shared publicationsSantwana Sagnika
School of Computer Engineering, KIIT-Deemed to be University, Bhubaneswar, 751024, Odisha, India
2 shared publicationsSatyabrata Roy
Department of Computer Science and Engineering, Manipal University Jaipur, Jaipur, 303007, Rajasthan, India
2 shared publications
Latest Publications
- Open Access
- 15 September 2026
- Published by Springer Nature in Scientific Reports
ViT-ConvGDNet: a vision transformer–MobileNet guided decoder network for robust copy-move forgery detection and localization
Copy-move forgery is a common form of image manipulation where a portion of an image is copied and pasted back into the image. This is especially difficult to detect when the forgery has been done on copied areas that have undergone post-processing operations, e.g. rotation, scaling, blurring etc. We propose a new encoder-decoder framework named ViT-ConvGDNet, which integrates the global contextual strengths of Vision Transformers with feature extraction strengths of convolutional operations in MobileNet. Sobel edge detection is added to the encoder to improve the level of awareness of the boundary and sharpness of features. Also, there is Atrous Spatial Pyramid Pooling (ASPP) to obtain multi-scale contextual data that are necessary to accurately perform localization. A layer-wise weighted loss mechanism controls the decoding process, which uses a custom mixture of loss functions at every decoder layer to improve the prediction accuracy. ViT-ConvGDNet makes use of patch-based self-attention mechanisms and is effective at learning long-range dependencies and adapts effectively to images of differing scales and complexity. The performance of the model is better as shown by extensive evaluations on several benchmark datasets such as MICC-F600, MICC-F2000, IMD, Coverage, CoMoFoD, Ardizzone, GRIP, CASIA, and USC-ISI. It has been experimentally demonstrated that ViT-ConvGDNet is more effective than various current deep learning methods and provides a rigorous and scalable solution to problematic copy-move forgery detection and localization.
- 15 September 2026
- Published by Springer Nature in Scientific Reports
Open AccessViT-ConvGDNet: a vision transformer–MobileNet guided decoder network for robust copy-move forgery detection and localization
Copy-move forgery is a common form of image manipulation where a portion of an image is copied and pasted back into the image. This is especially difficult to detect when the forgery has been done on copied areas that have undergone post-processing operations, e.g. rotation, scaling, blurring etc. We propose a new encoder-decoder framework named ViT-ConvGDNet, which integrates the global contextual strengths of Vision Transformers with feature extraction strengths of convolutional operations in MobileNet. Sobel edge detection is added to the encoder to improve the level of awareness of the boundary and sharpness of features. Also, there is Atrous Spatial Pyramid Pooling (ASPP) to obtain multi-scale contextual data that are necessary to accurately perform localization. A layer-wise weighted loss mechanism controls the decoding process, which uses a custom mixture of loss functions at every decoder layer to improve the prediction accuracy. ViT-ConvGDNet makes use of patch-based self-attention mechanisms and is effective at learning long-range dependencies and adapts effectively to images of differing scales and complexity. The performance of the model is better as shown by extensive evaluations on several benchmark datasets such as MICC-F600, MICC-F2000, IMD, Coverage, CoMoFoD, Ardizzone, GRIP, CASIA, and USC-ISI. It has been experimentally demonstrated that ViT-ConvGDNet is more effective than various current deep learning methods and provides a rigorous and scalable solution to problematic copy-move forgery detection and localization.
- Open Access
- 13 April 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1)
Four-port MIMO antenna loaded with FSS in FR2 band for wireless applications
This work reports a four-port MIMO antenna integrated with a novel wideband frequency-selective surface (FSS). The proposed MIMO antenna is fabricated on Rogers-dielectric with permittivity 2.20 and thickness of 0.254 mm with an overall area of 30 × 36 mm2. The radiating patch is achieved by merging three ring-hexagonal geometries, which are fed by an asymmetric tapered transmission line with a partial-rectangular ground printed on another surface with a pair of rectangular slots and an open-ended slot beneath the feedline. A novel de-coupling structure achieves isolation of more than 10.0 dB in MIMO configuration with a pair of radiating elements placed mirrored-adjacent sequence and the other pair placed oppositely. A novel 7 × 7 frequency-selective-surface of dimension 56 mm × 56 mm with an operating bandwidth of 10–43.30 GHz is also fabricated on the same substrate as an antenna and is placed 10.0 mm apart from the MIMO-antenna, which enhances the average peak-realized-gain by 8.0 dBi. The MIMO antenna achieves − 10.0 dB bandwidth of 3.45–7.25 GHz (Band-A) and 19.70–82.10 GHz (Band-B). The MIMO-FSS antenna achieves better diversity parameter with ECCM−FSS\(\:\le\:\) 0.02 (Band-A), 0.01 (Band-B), DGM−FSS\(\:>\) 9.998dB (Band-A), 9.996 dB (Band-B), TARCM−FSS\(\:\le\:\) − 4.0 dB (Band-A), − 6.0 dB (Band-B), CCLM−FSS\(\:\le\:\)0.18 b/s/Hz (Band-A), 0.25 b/s/Hz (Band-B), MEGM−FSS\(\:\cong\:\) − 3.0 dB (Band-A and Band-B), and MEG (porta-porta)\(\:\cong\:\)0.0 dB (Band-A and Band-B). The SARM−FSS values at 3.50 GHz, 5.90 GHz, 24.0 GHz, 28.0 GHz, 38.0 GHz, 39.0 GHz, 41.0 GHz, 43.0 GHz, 47.0 GHz and 60.0 GHz are below threshold value of 1.60 W/Kg at 1 g of the tissue. The bending analysis at 15°, 30°, and 45° of the MIMO-FSS antenna matches the − 10.0 dB bandwidth at 0-degree bending, suggesting application in flexible electronics. The antenna also maintains stable 2D radiation patterns in principal planes at a lower frequency of 3.50 GHz & 5.90 GHz, and is acceptable in the remaining higher frequency application bands. The MIMO proposed antenna loaded with FSS also records the enhancement of peak realized gain by 5.85 dBi in Band 2.- 13 April 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1)
Open AccessFour-port MIMO antenna loaded with FSS in FR2 band for wireless applications
This work reports a four-port MIMO antenna integrated with a novel wideband frequency-selective surface (FSS). The proposed MIMO antenna is fabricated on Rogers-dielectric with permittivity 2.20 and thickness of 0.254 mm with an overall area of 30 × 36 mm2. The radiating patch is achieved by merging three ring-hexagonal geometries, which are fed by an asymmetric tapered transmission line with a partial-rectangular ground printed on another surface with a pair of rectangular slots and an open-ended slot beneath the feedline. A novel de-coupling structure achieves isolation of more than 10.0 dB in MIMO configuration with a pair of radiating elements placed mirrored-adjacent sequence and the other pair placed oppositely. A novel 7 × 7 frequency-selective-surface of dimension 56 mm × 56 mm with an operating bandwidth of 10–43.30 GHz is also fabricated on the same substrate as an antenna and is placed 10.0 mm apart from the MIMO-antenna, which enhances the average peak-realized-gain by 8.0 dBi. The MIMO antenna achieves − 10.0 dB bandwidth of 3.45–7.25 GHz (Band-A) and 19.70–82.10 GHz (Band-B). The MIMO-FSS antenna achieves better diversity parameter with ECCM−FSS\(\:\le\:\) 0.02 (Band-A), 0.01 (Band-B), DGM−FSS\(\:>\) 9.998dB (Band-A), 9.996 dB (Band-B), TARCM−FSS\(\:\le\:\) − 4.0 dB (Band-A), − 6.0 dB (Band-B), CCLM−FSS\(\:\le\:\)0.18 b/s/Hz (Band-A), 0.25 b/s/Hz (Band-B), MEGM−FSS\(\:\cong\:\) − 3.0 dB (Band-A and Band-B), and MEG (porta-porta)\(\:\cong\:\)0.0 dB (Band-A and Band-B). The SARM−FSS values at 3.50 GHz, 5.90 GHz, 24.0 GHz, 28.0 GHz, 38.0 GHz, 39.0 GHz, 41.0 GHz, 43.0 GHz, 47.0 GHz and 60.0 GHz are below threshold value of 1.60 W/Kg at 1 g of the tissue. The bending analysis at 15°, 30°, and 45° of the MIMO-FSS antenna matches the − 10.0 dB bandwidth at 0-degree bending, suggesting application in flexible electronics. The antenna also maintains stable 2D radiation patterns in principal planes at a lower frequency of 3.50 GHz & 5.90 GHz, and is acceptable in the remaining higher frequency application bands. The MIMO proposed antenna loaded with FSS also records the enhancement of peak realized gain by 5.85 dBi in Band 2. - Open Access
- 11 April 2026
- Published by Springer Nature in BioData Mining
An efficient hybrid federated learning framework for pneumonia diagnosis with proximal optimization and parameter-efficient adaptation
- 11 April 2026
- Published by Springer Nature in BioData Mining
Open AccessAn efficient hybrid federated learning framework for pneumonia diagnosis with proximal optimization and parameter-efficient adaptation
- Open Access
- 9 April 2026
- Published by Springer Nature in Discover Artificial Intelligence
- Vol. 6 (1)
TSM-Transformer: a hybrid EfficientNetV2-Lite and temporal shift model for facial expression recognition in videos
- 9 April 2026
- Published by Springer Nature in Discover Artificial Intelligence
- Vol. 6 (1)
Open AccessTSM-Transformer: a hybrid EfficientNetV2-Lite and temporal shift model for facial expression recognition in videos
- Open Access
- 19 February 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1) , 6978
Correction: 3D FusionNet for synthetic CT based lung cancer segmentation
- 19 February 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1) , 6978
Open AccessCorrection: 3D FusionNet for synthetic CT based lung cancer segmentation
- Open Access
- 19 January 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1)
Combining predictive accuracy and interpretability: a data-driven approach to telecom churn analysis
Acquiring consumers is a challenge in the telecom industry since it reduces profits and slows growth. Customer churn prediction remains a pivotal challenge for telecommunications companies seeking to retain subscribers and sustain profitability in highly competitive markets. This study introduces XCL-Churn, an explainable ensemble learning framework that integrates XGBoost, CatBoost, and LightGBM within a soft-voting meta-architecture to achieve robust and interpretable churn prediction. The proposed framework leverages a comprehensive data preprocessing pipeline involving iterative Bayesian Ridge imputation for missing data, multi-stage scaling (Robust, Standard, and Min–Max), and feature refinement through a hybrid Boruta–Random Forest (RF) strategy. Class imbalance is addressed using the Synthetic Minority Oversampling Technique (SMOTE), ensuring equitable learning across churn and non-churn classes. The model is trained and validated on the large-scale Expresso Churn Prediction Challenge dataset from Zindi, encompassing approximately two million customer records with heterogeneous demographic, behavioral, and financial attributes. The novelty of this work lies in its holistic integration of advanced data imputation, multi-stage scaling, and hybrid feature selection within an explainable soft-voting ensemble that merges three state-of-the-art gradient boosting models. Experimental results demonstrate that the XCL-Churn ensemble achieves superior predictive performance—97.44% accuracy, 93.82% precision, 87.82% recall, and 91.25% F1-score—while maintaining computational efficiency with a reduced training time compared to traditional stacking and blending strategies. Furthermore, the incorporation of Explainable AI (XAI) techniques—LIME for local interpretability and SHAP for global feature importance—enhances transparency by elucidating the key behavioral and financial indicators driving customer churn decisions.
- 19 January 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1)
Open AccessCombining predictive accuracy and interpretability: a data-driven approach to telecom churn analysis
Acquiring consumers is a challenge in the telecom industry since it reduces profits and slows growth. Customer churn prediction remains a pivotal challenge for telecommunications companies seeking to retain subscribers and sustain profitability in highly competitive markets. This study introduces XCL-Churn, an explainable ensemble learning framework that integrates XGBoost, CatBoost, and LightGBM within a soft-voting meta-architecture to achieve robust and interpretable churn prediction. The proposed framework leverages a comprehensive data preprocessing pipeline involving iterative Bayesian Ridge imputation for missing data, multi-stage scaling (Robust, Standard, and Min–Max), and feature refinement through a hybrid Boruta–Random Forest (RF) strategy. Class imbalance is addressed using the Synthetic Minority Oversampling Technique (SMOTE), ensuring equitable learning across churn and non-churn classes. The model is trained and validated on the large-scale Expresso Churn Prediction Challenge dataset from Zindi, encompassing approximately two million customer records with heterogeneous demographic, behavioral, and financial attributes. The novelty of this work lies in its holistic integration of advanced data imputation, multi-stage scaling, and hybrid feature selection within an explainable soft-voting ensemble that merges three state-of-the-art gradient boosting models. Experimental results demonstrate that the XCL-Churn ensemble achieves superior predictive performance—97.44% accuracy, 93.82% precision, 87.82% recall, and 91.25% F1-score—while maintaining computational efficiency with a reduced training time compared to traditional stacking and blending strategies. Furthermore, the incorporation of Explainable AI (XAI) techniques—LIME for local interpretability and SHAP for global feature importance—enhances transparency by elucidating the key behavioral and financial indicators driving customer churn decisions.
- Open Access
- 29 December 2025
- Published by Springer Nature in Journal of Cloud Computing
- Vol. 14 (1) , 78
An access control system for cloud-based healthcare systems driven by blockchain
- 29 December 2025
- Published by Springer Nature in Journal of Cloud Computing
- Vol. 14 (1) , 78
Open AccessAn access control system for cloud-based healthcare systems driven by blockchain
- Open Access
- 26 December 2025
- Published by Springer Nature in Discover Applied Sciences
- Vol. 8 (2)
Secure and scalable smart grid IoT communication through quantum key distribution, homomorphic encryption, and federated learning
- 26 December 2025
- Published by Springer Nature in Discover Applied Sciences
- Vol. 8 (2)
Open AccessSecure and scalable smart grid IoT communication through quantum key distribution, homomorphic encryption, and federated learning
- Open Access
- 3 December 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1)
3D FusionNet for synthetic CT based lung cancer segmentation
Interpretation of three-dimensional medical scans continues to be a demanding task in computer-assisted diagnostic systems. In this study we present a segmentation framework specifically designed for lung CT images. Since lung cancer is one of the most fatal cancers worldwide and the demand for better detection tools remains pressing. To address this issue our proposed work combined the Deep Convolutional Generative Adversarial Networks (DCGAN) with the 3D-TDUnet++ architecture. The use of DCGAN allowed us to generate additional CT samples. Which results in reducing the problem of insufficient annotated data and improving the robustness of the training set. In this experiment the Chest CT-Scan images Dataset is used which is publicly available in KAGGLE. Which is being further enriched with synthetic images created by DCGAN. While being trained on this augmented dataset the proposed 3D-FusionNet model achieved superior performance in cancer detection. Which shows higher accuracy, sensitivity and specificity than other conventional approaches. The integration of DCGAN with 3D-TDUnet++ together with Non-Local Feature Aggregation (NLFa) made the system promising for clinical environments where limited data often restricts the application of deep learning models. Quantitative evaluation reveals that the proposed 3D-FusionNet achieves a Dice coefficient of 88.94%, F1-score of 88.94% and an accuracy of 93.37%. Outperforming the benchmark models such as DenseNet, ResNet and MDDNet-ASPP. This hybrid design leverages the strengths of generative augmentation and non-local attention to improve both segmentation precision and clinical robustness. Thus the proposed 3D-FusionNet model demonstrates how deep generative modeling and volumetric segmentation can be integrated into a cohesive machine vision system. Which is capable of scalable and efficient deployment in real-world diagnostic pipelines.
- 3 December 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1)
Open Access3D FusionNet for synthetic CT based lung cancer segmentation
Interpretation of three-dimensional medical scans continues to be a demanding task in computer-assisted diagnostic systems. In this study we present a segmentation framework specifically designed for lung CT images. Since lung cancer is one of the most fatal cancers worldwide and the demand for better detection tools remains pressing. To address this issue our proposed work combined the Deep Convolutional Generative Adversarial Networks (DCGAN) with the 3D-TDUnet++ architecture. The use of DCGAN allowed us to generate additional CT samples. Which results in reducing the problem of insufficient annotated data and improving the robustness of the training set. In this experiment the Chest CT-Scan images Dataset is used which is publicly available in KAGGLE. Which is being further enriched with synthetic images created by DCGAN. While being trained on this augmented dataset the proposed 3D-FusionNet model achieved superior performance in cancer detection. Which shows higher accuracy, sensitivity and specificity than other conventional approaches. The integration of DCGAN with 3D-TDUnet++ together with Non-Local Feature Aggregation (NLFa) made the system promising for clinical environments where limited data often restricts the application of deep learning models. Quantitative evaluation reveals that the proposed 3D-FusionNet achieves a Dice coefficient of 88.94%, F1-score of 88.94% and an accuracy of 93.37%. Outperforming the benchmark models such as DenseNet, ResNet and MDDNet-ASPP. This hybrid design leverages the strengths of generative augmentation and non-local attention to improve both segmentation precision and clinical robustness. Thus the proposed 3D-FusionNet model demonstrates how deep generative modeling and volumetric segmentation can be integrated into a cohesive machine vision system. Which is capable of scalable and efficient deployment in real-world diagnostic pipelines.
- Top Cited PapersOpen Access
- 3 November 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 1-23
Fraudulent account detection in social media using hybrid deep transformer model and hyperparameter optimization
The high rate of social media development has triggered a high rate of fake accounts, which are a great risk to the privacy of users and the integrity of the platform. These malicious accounts are hard to detect because user activity data is highly imbalanced, dimensional, and sequential. The emergence of fake profiles on social media endangers the privacy and trust of social media users. It is difficult to detect such accounts because of high-dimensional, highly sequential, and imbalanced user behavior data. Current techniques tend to miss out on the complicated activity patterns or even overfit, which is why a strong, scalable, and precise model of social media fraud detection is required. This study suggests a new deep learning architecture that entails a Temporal Convolutional Network (TCN) with Generative Adversarial Network (GAN)-based data augmentation to generate minority classes, and Autoencoder-based feature extraction to reduce dimensionality. The Seagull Optimization Algorithm (SOA), which is a metaheuristic algorithm, is used to optimize hyperparameters by balancing efficiency and speed of convergence in global search. The framework is tested on benchmark datasets (Cresci-2017 and TwiBot-22) and compared to the state-of-the-art models. It has been shown in experiments that the suggested TCN-GAN-SOA framework performs better, with ROC-AUC scores of 0.96 on Cresci-2017 and 0.95 on TwiBot-22, and a higher precision-recall value and better F1-scores. In addition, computational efficiency can be verified by the runtime analysis; case studies prove the framework’s strength when handling various situations of fraudulent behaviors. The given solution offers a scalable, reliable, and accurate methodology of detecting social media fraud based on the combination of sophisticated sequence modeling, realistic data augmentation, and hyperparameter optimization.
- 3 November 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 1-23
Top Cited PapersOpen AccessFraudulent account detection in social media using hybrid deep transformer model and hyperparameter optimization
The high rate of social media development has triggered a high rate of fake accounts, which are a great risk to the privacy of users and the integrity of the platform. These malicious accounts are hard to detect because user activity data is highly imbalanced, dimensional, and sequential. The emergence of fake profiles on social media endangers the privacy and trust of social media users. It is difficult to detect such accounts because of high-dimensional, highly sequential, and imbalanced user behavior data. Current techniques tend to miss out on the complicated activity patterns or even overfit, which is why a strong, scalable, and precise model of social media fraud detection is required. This study suggests a new deep learning architecture that entails a Temporal Convolutional Network (TCN) with Generative Adversarial Network (GAN)-based data augmentation to generate minority classes, and Autoencoder-based feature extraction to reduce dimensionality. The Seagull Optimization Algorithm (SOA), which is a metaheuristic algorithm, is used to optimize hyperparameters by balancing efficiency and speed of convergence in global search. The framework is tested on benchmark datasets (Cresci-2017 and TwiBot-22) and compared to the state-of-the-art models. It has been shown in experiments that the suggested TCN-GAN-SOA framework performs better, with ROC-AUC scores of 0.96 on Cresci-2017 and 0.95 on TwiBot-22, and a higher precision-recall value and better F1-scores. In addition, computational efficiency can be verified by the runtime analysis; case studies prove the framework’s strength when handling various situations of fraudulent behaviors. The given solution offers a scalable, reliable, and accurate methodology of detecting social media fraud based on the combination of sophisticated sequence modeling, realistic data augmentation, and hyperparameter optimization.