Shilpa Ankalaki
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Shilpa Ankalaki has a total of 32 co-authors. Here are the top co-authors by number of shared publications:
- Geetabai S. Hukkeri
Manipal Institute of Technology Bengaluru, Manipal Academy of Higher Education, Manipal, India
8 shared publications - Rayangouda H. Goudar
Visvesvaraya Technological University
3 shared publications - Anjanabhargavi Kulkarni
Department of Computer Science and Engineering, Visvesvaraya Technological University, Belagavi, India
2 shared publications - Dhananjaya. G. M.
Visvesvaraya Technological University,Dept of Computer Science and Engineering,Belagavi,Karnataka,India
2 shared publications Aarya Shah
Manipal Institute of Technology Bengaluru, Manipal Academy of Higher Education,Manipal,Karnataka,India,576104
1 shared publicationAkella S Narasimha Raju
Department of Computing Technologies, School of Computing, College of Engineering Technology, SRM Institute of Science and Technology, Kattakalathur, Chennai, Tamil Nadu, India
1 shared publication
Latest Publications
- Open Access
- 26 August 2026
- Published by Springer Nature in Discover Applied Sciences
A systematic survey of digital twins for endometriosis prediction
Endometriosis is a chronic condition of the female reproductive system that demands sustained clinical attention. It can affect the fallopian tubes, ovaries, pelvic organs, and uterine tissues, often leading to persistent pain, fertility challenges, and reduced quality of life. Although the condition is estimated to impact nearly 10% of women and girls of reproductive age worldwide, timely diagnosis remains difficult. Most patients experience a delay of 6–9 years before receiving confirmation, largely due to non-specific symptoms, limited clinical suspicion, and reliance on laparoscopy as the definitive diagnostic method. This review examines the evolution of predictive approaches in healthcare, emphasizing the importance of early identification of endometriosis. It explores how artificial intelligence and machine learning have been applied to enhance diagnostic accuracy, with models such as random forest, XGBoost, and convolutional neural networks demonstrating promising performance, achieving AUC values between 0.70 and 0.95 across recent studies. Following PRISMA-based screening, 42 studies published between 2019 and 2025 were included, focusing on prediction, machine learning, and digital-health applications. The synthesis highlights growing interest in integrating Digital Twin technology with explainable AI. Such integration has the potential to shift endometriosis care from reactive treatment to proactive, personalized prediction, enabling non-invasive early detection and more patient-specific clinical decision-making.
- 26 August 2026
- Published by Springer Nature in Discover Applied Sciences
Open AccessA systematic survey of digital twins for endometriosis prediction
Endometriosis is a chronic condition of the female reproductive system that demands sustained clinical attention. It can affect the fallopian tubes, ovaries, pelvic organs, and uterine tissues, often leading to persistent pain, fertility challenges, and reduced quality of life. Although the condition is estimated to impact nearly 10% of women and girls of reproductive age worldwide, timely diagnosis remains difficult. Most patients experience a delay of 6–9 years before receiving confirmation, largely due to non-specific symptoms, limited clinical suspicion, and reliance on laparoscopy as the definitive diagnostic method. This review examines the evolution of predictive approaches in healthcare, emphasizing the importance of early identification of endometriosis. It explores how artificial intelligence and machine learning have been applied to enhance diagnostic accuracy, with models such as random forest, XGBoost, and convolutional neural networks demonstrating promising performance, achieving AUC values between 0.70 and 0.95 across recent studies. Following PRISMA-based screening, 42 studies published between 2019 and 2025 were included, focusing on prediction, machine learning, and digital-health applications. The synthesis highlights growing interest in integrating Digital Twin technology with explainable AI. Such integration has the potential to shift endometriosis care from reactive treatment to proactive, personalized prediction, enabling non-invasive early detection and more patient-specific clinical decision-making.
- Open Access
- 18 May 2026
- Published by Springer Nature in Discover Applied Sciences
- Vol. 8 (7)
Adaptive resource management in 6G mmWave networks using K-means clustering and Q-learning for efficient bandwidth optimization
This paper proposes a novel hierarchical machine learning architecture for scalable resource management in 6G millimeter-wave networks, focusing on overcoming the curse of dimensionality in reinforcement learning for ultra-dense deployments. This work achieves a synergistic integration of K-means clustering with Q-learning and develop a two-stage optimization framework that adapts dynamically to user mobility and variations in channels. Unlike the existing monolithic or multi-agent RL approaches, first groups with similar-bandwidth demand users according to channel conditions by applying K-means, reducing the state-action space from U users to K clusters (K < < U). Then, dedicated Q-learning agents per cluster perform joint channel selection, bandwidth allocation, power control, and beamwidth adaptation under a unified multi-objective reward function. Extensive simulations using NS-3 are performed with NYUSIM channel models and demonstrate exceptional performance gains: achieving 150 Mbps mean bandwidth optimization 99% improvement over conventional 5G methods-while maintaining robust scalability, where improve the effective signal quality by intelligently managing interference, resulting in a higher Signal-to-Interference-plus-Noise Ratio (SINR) and more efficient bandwidth utilization improves from 52 dB to 60 dB while increasing the device density from 10 to 100 devices/km². This framework provides a practical solution to the scalability-efficiency tradeoff in RL-based wireless resource management and presents considerable enhancements in network capacity, spectral efficiency, and adaptive channel allocation for next-generation 6G systems.
- 18 May 2026
- Published by Springer Nature in Discover Applied Sciences
- Vol. 8 (7)
Open AccessAdaptive resource management in 6G mmWave networks using K-means clustering and Q-learning for efficient bandwidth optimization
This paper proposes a novel hierarchical machine learning architecture for scalable resource management in 6G millimeter-wave networks, focusing on overcoming the curse of dimensionality in reinforcement learning for ultra-dense deployments. This work achieves a synergistic integration of K-means clustering with Q-learning and develop a two-stage optimization framework that adapts dynamically to user mobility and variations in channels. Unlike the existing monolithic or multi-agent RL approaches, first groups with similar-bandwidth demand users according to channel conditions by applying K-means, reducing the state-action space from U users to K clusters (K < < U). Then, dedicated Q-learning agents per cluster perform joint channel selection, bandwidth allocation, power control, and beamwidth adaptation under a unified multi-objective reward function. Extensive simulations using NS-3 are performed with NYUSIM channel models and demonstrate exceptional performance gains: achieving 150 Mbps mean bandwidth optimization 99% improvement over conventional 5G methods-while maintaining robust scalability, where improve the effective signal quality by intelligently managing interference, resulting in a higher Signal-to-Interference-plus-Noise Ratio (SINR) and more efficient bandwidth utilization improves from 52 dB to 60 dB while increasing the device density from 10 to 100 devices/km². This framework provides a practical solution to the scalability-efficiency tradeoff in RL-based wireless resource management and presents considerable enhancements in network capacity, spectral efficiency, and adaptive channel allocation for next-generation 6G systems.
- Open Access
- 31 March 2026
- Published by Springer Nature in Archives of Computational Methods in Engineering
- Vol. 33 (5) , 7369-7396
Advancing Healthcare Using Quantum Computing and Federated Learning: Challenges and Future Perspectives
- 31 March 2026
- Published by Springer Nature in Archives of Computational Methods in Engineering
- Vol. 33 (5) , 7369-7396
Open AccessAdvancing Healthcare Using Quantum Computing and Federated Learning: Challenges and Future Perspectives
- Open Access
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 1
Domain-Invariant Vision-Language Models for Fashion E-Commerce Retrieval: Challenges, Benchmarks and Industry Perspectives
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 1
Open AccessDomain-Invariant Vision-Language Models for Fashion E-Commerce Retrieval: Challenges, Benchmarks and Industry Perspectives
- Open Access
- 5 December 2025
- Published by Springer Nature in Discover Applied Sciences
- Vol. 8 (1)
Dynamic 6G propagation modeling with reconfigurable intelligent surfaces and machine learning for urban rural highway and hilly mobility scenarios
Conventional propagation models struggle to accurately characterize the extreme mobility and complex environmental challenges anticipated in 6G networks. These include severe Doppler shifts on highways (120–500 km/h), profound urban non-line-of-sight (NLoS) blockages exceeding 140 dB at 200 m, rural energy constraints leading to 30–50% reductions in transmit power, and irregular hilly terrains with path loss exponents as high as 3.8. To address this gap, we propose an integrated propagation framework that combines classical physical modeling, Reconfigurable Intelligent Surface (RIS) enhancement, and AI-driven optimization. Our implementation utilizes the Close-In (CI) path loss model with environment-specific exponents and Rician fading with scenario-dependent K-factors. To mitigate high attenuation and fading, the framework incorporates 64-element RIS structures, which provide an SNR improvement of up to + 18 dB, and employs reinforcement learning for dynamic beamforming, adaptive phase management, and anticipatory resource allocation. Validation across four mobility scenarios demonstrates significant performance gains: without RIS, highway SNR remains above 10 dB up to 500 m, but urban SNR falls below 0 dB at ~ 100 m and hilly conditions degrade to − 5 dB at 300 m. With RIS, urban and hilly environments maintain an SNR > 10 dB up to 250 m, while highway and rural scenarios achieve over 15 dB SNR at 700 m. Throughput is enhanced from below 100 Mbps at distances beyond 100 m in baseline urban/hilly conditions to over 500 Mbps at 300 m. Furthermore, AI-powered beam alignment reduces handover failures by 40% under high-speed highway mobility.
- 5 December 2025
- Published by Springer Nature in Discover Applied Sciences
- Vol. 8 (1)
Open AccessDynamic 6G propagation modeling with reconfigurable intelligent surfaces and machine learning for urban rural highway and hilly mobility scenarios
Conventional propagation models struggle to accurately characterize the extreme mobility and complex environmental challenges anticipated in 6G networks. These include severe Doppler shifts on highways (120–500 km/h), profound urban non-line-of-sight (NLoS) blockages exceeding 140 dB at 200 m, rural energy constraints leading to 30–50% reductions in transmit power, and irregular hilly terrains with path loss exponents as high as 3.8. To address this gap, we propose an integrated propagation framework that combines classical physical modeling, Reconfigurable Intelligent Surface (RIS) enhancement, and AI-driven optimization. Our implementation utilizes the Close-In (CI) path loss model with environment-specific exponents and Rician fading with scenario-dependent K-factors. To mitigate high attenuation and fading, the framework incorporates 64-element RIS structures, which provide an SNR improvement of up to + 18 dB, and employs reinforcement learning for dynamic beamforming, adaptive phase management, and anticipatory resource allocation. Validation across four mobility scenarios demonstrates significant performance gains: without RIS, highway SNR remains above 10 dB up to 500 m, but urban SNR falls below 0 dB at ~ 100 m and hilly conditions degrade to − 5 dB at 300 m. With RIS, urban and hilly environments maintain an SNR > 10 dB up to 250 m, while highway and rural scenarios achieve over 15 dB SNR at 700 m. Throughput is enhanced from below 100 Mbps at distances beyond 100 m in baseline urban/hilly conditions to over 500 Mbps at 300 m. Furthermore, AI-powered beam alignment reduces handover failures by 40% under high-speed highway mobility.
- Open Access
- 23 October 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 1-27
Automating wastewater characteristic parameter quantitation using neural architecture search in AutoML systems on spectral reflectance data
Wastewater (WW) analyses are required to establish optimal operation of treatment facilities, and rapid treatment assessment can greatly improve operational efficiency. WWTPs function through complex biochemical processes that exhibit high variability and are difficult to predict. This work presents the first in-depth study towards an original application of Auto Machine Learning (AutoML) models to accurately predict quality parameters of wastewater. This study aims to demonstrate the efficiency of neural network (NN) regression models built by applying the Neural Architectural Space (NAS) applying different search algorithms such as random search, grid search, Bayesian Optimization, and Hyperband search to predict the concentrations of various wastewater (WW) parameters such as biochemical oxygen demand (BOD), chemical oxygen demand (COD), ammonia (NH 3 -N), total dissolved solids (TDS), total alkalinity (TA), and total hardness (TH). The models are trained on data based on WW spectral reflectance data in the visible-near infrared range (400–2000 nm). The input variables consisted of spectral reflectance intensity of wastewater, while the output variables consisted of levels of six wastewater (WW) parameters, namely BOD, COD, NH 3 -N, TDS, TA, and TH. Different search algorithms were employed to identify the neural network (NN) architectures that provided single and multiple target variable (two to six target variables) predictions to the analysis. Experimental results indicate that the NN architecture optimized via Bayesian Optimization outperforms others, achieving the lowest mean absolute error (MAE) and a high coefficient of determination (R 2 ). For single-target predictions on the validation set, the model attained R 2 values of 0.9770 for BOD, 0.9860 for COD, 0.9800 for NH 3 -N, 0.9776 for TDS, 0.9847 for TA, and 0.9639 for TH. For two target predictions on the validation set, the model obtained R 2 values of 0.988 for BOD and 0.993 for COD. Similarly R 2 for predictions for three target variables are 0.961 for BOD, 0.962 for COD and 0.946 for NH3-N; for four target predictions 0.972 for BOD, 0.970 for COD, 0.955 for NH 3 -N and 0.972 for TDS; for five predictions 0.966 for BOD, 0.966 for COD, 0.951 for NH 3 -N, 0.966 for TDS and 0.966 for TA; and for six target predictions 0.987 for BOD, 0.99 for COD, 0.982 for NH 3 -N, 0.99 for TDS, 0.99 for TA and 0.718 for TH. Bayesian optimization search-based NN predicted the TH value with an R 2 of 0.718. The Hyperband search NN obtained better results for TH prediction during six target predictions i.e. 0.9772 for BOD, 0.976 for COD, 0.975 for NH 3 -N, 0.976 for TDS, 0.976 for TA, and 0.899 for TH. This work automates the neural network architecture optimization process using NAS techniques. These methods systematically explore the hyperparameter space, removing the need for manual trial-and-error tuning across different datasets and target prediction tasks (both single and multi-target). The resulting neural network models are further interpreted using the LIME Explainable Artificial Intelligence (XAI) method.
- 23 October 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 1-27
Open AccessAutomating wastewater characteristic parameter quantitation using neural architecture search in AutoML systems on spectral reflectance data
Wastewater (WW) analyses are required to establish optimal operation of treatment facilities, and rapid treatment assessment can greatly improve operational efficiency. WWTPs function through complex biochemical processes that exhibit high variability and are difficult to predict. This work presents the first in-depth study towards an original application of Auto Machine Learning (AutoML) models to accurately predict quality parameters of wastewater. This study aims to demonstrate the efficiency of neural network (NN) regression models built by applying the Neural Architectural Space (NAS) applying different search algorithms such as random search, grid search, Bayesian Optimization, and Hyperband search to predict the concentrations of various wastewater (WW) parameters such as biochemical oxygen demand (BOD), chemical oxygen demand (COD), ammonia (NH 3 -N), total dissolved solids (TDS), total alkalinity (TA), and total hardness (TH). The models are trained on data based on WW spectral reflectance data in the visible-near infrared range (400–2000 nm). The input variables consisted of spectral reflectance intensity of wastewater, while the output variables consisted of levels of six wastewater (WW) parameters, namely BOD, COD, NH 3 -N, TDS, TA, and TH. Different search algorithms were employed to identify the neural network (NN) architectures that provided single and multiple target variable (two to six target variables) predictions to the analysis. Experimental results indicate that the NN architecture optimized via Bayesian Optimization outperforms others, achieving the lowest mean absolute error (MAE) and a high coefficient of determination (R 2 ). For single-target predictions on the validation set, the model attained R 2 values of 0.9770 for BOD, 0.9860 for COD, 0.9800 for NH 3 -N, 0.9776 for TDS, 0.9847 for TA, and 0.9639 for TH. For two target predictions on the validation set, the model obtained R 2 values of 0.988 for BOD and 0.993 for COD. Similarly R 2 for predictions for three target variables are 0.961 for BOD, 0.962 for COD and 0.946 for NH3-N; for four target predictions 0.972 for BOD, 0.970 for COD, 0.955 for NH 3 -N and 0.972 for TDS; for five predictions 0.966 for BOD, 0.966 for COD, 0.951 for NH 3 -N, 0.966 for TDS and 0.966 for TA; and for six target predictions 0.987 for BOD, 0.99 for COD, 0.982 for NH 3 -N, 0.99 for TDS, 0.99 for TA and 0.718 for TH. Bayesian optimization search-based NN predicted the TH value with an R 2 of 0.718. The Hyperband search NN obtained better results for TH prediction during six target predictions i.e. 0.9772 for BOD, 0.976 for COD, 0.975 for NH 3 -N, 0.976 for TDS, 0.976 for TA, and 0.899 for TH. This work automates the neural network architecture optimization process using NAS techniques. These methods systematically explore the hyperparameter space, removing the need for manual trial-and-error tuning across different datasets and target prediction tasks (both single and multi-target). The resulting neural network models are further interpreted using the LIME Explainable Artificial Intelligence (XAI) method.
- Open Access
- 14 October 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 35853
SpinachXAI-Rec: a multi-stage explainable AI framework for spinach freshness classification and consumer recommendation
Leafy vegetables such as spinach are among the most important components in a nutritious diet but are highly perishable and susceptible to premature spoilage. Traditional practices in determining freshness have been qualitative and time-consuming and have consistently led to defective consumption decisions with unintended consequences on human health. To this issue, we introduce SpinachXAI-Rec, a multistage framework that is enabled by AI and is capable of automating the classification of spinach freshness and providing consumer recommendations. This framework is based on understandable deep learning. To guarantee class balance and feature diversity, a dataset consisting of 4005 original images of three spinach varieties (Malabar, Red, and Water) was expanded to 12,000 images (2000 per class across six categories: fresh and non-fresh). We trained three CNN architectures, DenseNet121, ResNet50, and EfficientNetB0, on the Stage 1 augmented dataset. In performance, we saw DenseNet121 significantly outperform with 96% classification accuracy compared to ResNet50 (53%) and EfficientNetB0 (17%). Stage 2 improved representation of features by incorporating DenseNet121 embeddings and ViT-B/16 and Swin Transformer attention mechanisms. DenseNet121 + ViT-B/16 obtained an F1-score of 0.95, which was further optimised to 0.97 in Stage 3 using a multiclass SVM classifier. GradCAM++ and LIME were employed to incorporate interpretability during Stage 4. LIME provided transparent explanations of the significance of class-specific features, while GradCAM++ effectively highlighted disease-affected or spoilt regions. The most effective model (DenseNet121 + ViT + SVM) also obtained a Dice coefficient of 0.89 and an IoU of 0.82, which confirms the precision of localisation and segmentation. Finally, Stage 5 introduces a clinical recommender system that is based on rules and relates prediction confidence to real-world categories: Eatable, Eatable with Caution, or Not Eatable. This AI-driven recommendation assists food purveyors and consumers in making health-conscious, well-informed decisions. SpinachXAI-Rec is a significant advancement in the development of safer food systems, as it provides interpretable AI for the purpose of freshness validation and actionable consumption recommendations, thereby empowering both consumers and industry stakeholders.
- 14 October 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 35853
Open AccessSpinachXAI-Rec: a multi-stage explainable AI framework for spinach freshness classification and consumer recommendation
Leafy vegetables such as spinach are among the most important components in a nutritious diet but are highly perishable and susceptible to premature spoilage. Traditional practices in determining freshness have been qualitative and time-consuming and have consistently led to defective consumption decisions with unintended consequences on human health. To this issue, we introduce SpinachXAI-Rec, a multistage framework that is enabled by AI and is capable of automating the classification of spinach freshness and providing consumer recommendations. This framework is based on understandable deep learning. To guarantee class balance and feature diversity, a dataset consisting of 4005 original images of three spinach varieties (Malabar, Red, and Water) was expanded to 12,000 images (2000 per class across six categories: fresh and non-fresh). We trained three CNN architectures, DenseNet121, ResNet50, and EfficientNetB0, on the Stage 1 augmented dataset. In performance, we saw DenseNet121 significantly outperform with 96% classification accuracy compared to ResNet50 (53%) and EfficientNetB0 (17%). Stage 2 improved representation of features by incorporating DenseNet121 embeddings and ViT-B/16 and Swin Transformer attention mechanisms. DenseNet121 + ViT-B/16 obtained an F1-score of 0.95, which was further optimised to 0.97 in Stage 3 using a multiclass SVM classifier. GradCAM++ and LIME were employed to incorporate interpretability during Stage 4. LIME provided transparent explanations of the significance of class-specific features, while GradCAM++ effectively highlighted disease-affected or spoilt regions. The most effective model (DenseNet121 + ViT + SVM) also obtained a Dice coefficient of 0.89 and an IoU of 0.82, which confirms the precision of localisation and segmentation. Finally, Stage 5 introduces a clinical recommender system that is based on rules and relates prediction confidence to real-world categories: Eatable, Eatable with Caution, or Not Eatable. This AI-driven recommendation assists food purveyors and consumers in making health-conscious, well-informed decisions. SpinachXAI-Rec is a significant advancement in the development of safer food systems, as it provides interpretable AI for the purpose of freshness validation and actionable consumption recommendations, thereby empowering both consumers and industry stakeholders.
- Open Access
- 1 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. PP, 1
XAI-SkinCADx: A Six-Stage Explainable Deep Ensemble Framework for Skin Cancer Diagnosis and Risk-Based Clinical Recommendations
The fastest growing and deadliest cancer, skin cancer, requires diagnostic solutions that are accurate, understandable, and therapeutically actionable. Our six-stage deep ensemble diagnostic framework, XAI-SkinCADx, employs hybrid feature extraction, deep learning ensembles, understandable AI, and clinical recommendation creation to meet this need. The 2367 dermoscopy images belonging to nine distinct classes of the ISIC skin cancer dataset are used to train the framework. Adding 5,023 new images makes the dataset richer and better balanced across the classes. A comprehensive performance evaluation is ensured by using the dataset with an 80:20 train-data/test-data split. It begins with hand-designed feature extraction using GLCM-based dimensionality reduction and visualization and Local Binary Patterns (LBP) visualization. Three convolutional neural networks are then employed to extract specific spatial features: DV-25 (DenseNet201 + VGG16), DE-25 (DenseNet201 + EfficientNetB5), and DM-25 (DenseNet201 + MobileNetV2). Temporal patterns are described with bidirectional long-short-term memory (BiLSTM) and finally classified with multiclass SVM. The accuracy is 95.63%, precision is 96.2%, recall is 95.7%, F1 score is 95.9%, and area under curve is 0.97. We interpret it using Grad-CAM++ and LIME. Compared to Grad-CAM++, LIME had better localization accuracy (IoU: 0.78, Dice: 0.87) while Grad-CAM++ had worse accuracy (IoU: 0.70, Dice: 0.81). The best performing classes with LIME interpretability, Vascular Lesion, Seborrhoeic Keratosis, and Dermatofibroma, respectively, with Dice scores of 0.91, 0.89, and 0.87 respectively. For providing patients with personalized suggestions independent of other information, a dermatology-focused LIME-based recommendation platform was designed. The timely dermatology consultation and decreased clinical burden are led by the classification by the system of the disorders as low-risk, medium-risk, and high-risk. By integrating accuracy, clarity, and utility, XAI-SkinCADx creates a new benchmark for applying AI in the diagnosis of skin disease, enabling physicians to make well-informed and trusted decisions.
- 1 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. PP, 1
Open AccessXAI-SkinCADx: A Six-Stage Explainable Deep Ensemble Framework for Skin Cancer Diagnosis and Risk-Based Clinical Recommendations
The fastest growing and deadliest cancer, skin cancer, requires diagnostic solutions that are accurate, understandable, and therapeutically actionable. Our six-stage deep ensemble diagnostic framework, XAI-SkinCADx, employs hybrid feature extraction, deep learning ensembles, understandable AI, and clinical recommendation creation to meet this need. The 2367 dermoscopy images belonging to nine distinct classes of the ISIC skin cancer dataset are used to train the framework. Adding 5,023 new images makes the dataset richer and better balanced across the classes. A comprehensive performance evaluation is ensured by using the dataset with an 80:20 train-data/test-data split. It begins with hand-designed feature extraction using GLCM-based dimensionality reduction and visualization and Local Binary Patterns (LBP) visualization. Three convolutional neural networks are then employed to extract specific spatial features: DV-25 (DenseNet201 + VGG16), DE-25 (DenseNet201 + EfficientNetB5), and DM-25 (DenseNet201 + MobileNetV2). Temporal patterns are described with bidirectional long-short-term memory (BiLSTM) and finally classified with multiclass SVM. The accuracy is 95.63%, precision is 96.2%, recall is 95.7%, F1 score is 95.9%, and area under curve is 0.97. We interpret it using Grad-CAM++ and LIME. Compared to Grad-CAM++, LIME had better localization accuracy (IoU: 0.78, Dice: 0.87) while Grad-CAM++ had worse accuracy (IoU: 0.70, Dice: 0.81). The best performing classes with LIME interpretability, Vascular Lesion, Seborrhoeic Keratosis, and Dermatofibroma, respectively, with Dice scores of 0.91, 0.89, and 0.87 respectively. For providing patients with personalized suggestions independent of other information, a dermatology-focused LIME-based recommendation platform was designed. The timely dermatology consultation and decreased clinical burden are led by the classification by the system of the disorders as low-risk, medium-risk, and high-risk. By integrating accuracy, clarity, and utility, XAI-SkinCADx creates a new benchmark for applying AI in the diagnosis of skin disease, enabling physicians to make well-informed and trusted decisions.
- 3 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 113-118
Stock Price Prediction with Optimized Xgboost: a Comprehensive Benchmark Against Deep Learning Approaches
Historical medium-term stock price prediction remains a critical but challenging endeavor regardless of market turbulence, non-stationarity and high computational complexities of recent prediction models. In this work, a Bayesian-tuned Extreme Gradient Boosting (XGBoost) pipeline is advanced to be specifically suitable for 60-day ahead stock price prediction. This approach also includes inclusion of strict walk-forward validation, hardware-aware parallel computation and interpretable feature identification from SHAP values. Based on 14 years of NASDAQ history, the estimated optimal XGBoost model has produced an RMSE of 4.23 on the stock price ticker AAPL with a reduction in error by 34.78% over a baseline bi-directional LSTM model while training 9.1 times faster and using 89.3% less GPU time. Backtesting on simulations of trading resulted in a very high cumulative portfolio return in five years, beating deep learning and buy and hold baselines. The model maintained an excellent performance in financial, technology, healthcare, energy and consumer sectors even during high volatility periods like the COVID-19 crash. All these outcomes prove the potential of using optimized gradient boosting models in achieving high accuracy and interpretable solutions for real-time financial prediction and algorithmic trading.
- 3 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 113-118
Stock Price Prediction with Optimized Xgboost: a Comprehensive Benchmark Against Deep Learning Approaches
Historical medium-term stock price prediction remains a critical but challenging endeavor regardless of market turbulence, non-stationarity and high computational complexities of recent prediction models. In this work, a Bayesian-tuned Extreme Gradient Boosting (XGBoost) pipeline is advanced to be specifically suitable for 60-day ahead stock price prediction. This approach also includes inclusion of strict walk-forward validation, hardware-aware parallel computation and interpretable feature identification from SHAP values. Based on 14 years of NASDAQ history, the estimated optimal XGBoost model has produced an RMSE of 4.23 on the stock price ticker AAPL with a reduction in error by 34.78% over a baseline bi-directional LSTM model while training 9.1 times faster and using 89.3% less GPU time. Backtesting on simulations of trading resulted in a very high cumulative portfolio return in five years, beating deep learning and buy and hold baselines. The model maintained an excellent performance in financial, technology, healthcare, energy and consumer sectors even during high volatility periods like the COVID-19 crash. All these outcomes prove the potential of using optimized gradient boosting models in achieving high accuracy and interpretable solutions for real-time financial prediction and algorithmic trading.
- Top Cited PapersOpen Access
- 3 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 44662-44706
Cyber Attack Prediction: From Traditional Machine Learning to Generative Artificial Intelligence
The escalating sophistication of cyber threats poses significant risks to individuals, organizations, and nations. Cybercrime, encompassing activities like hacking and data breaches, has severe economic and societal consequences. In today’s interconnected world, robust cybersecurity measures are paramount to mitigate these risks and protect sensitive information. However, traditional security solutions struggle to keep pace with the evolving threat landscape. Artificial Intelligence (AI) offers a powerful arsenal of techniques to address these challenges. This paper explores the application of AI methods, including Machine Learning (ML), Deep Learning (DL), Natural Language Processing (NLP), Explainable AI (XAI), and Generative AI, in solving various cybersecurity problems. This paper presents a comprehensive analysis of AI techniques for enhancing cybersecurity. Key contributions include: i) A comparative study of ML and DL methods: Evaluating their accuracy, applicability, and suitability for various cybersecurity challenges. ii) An investigation into XAI approaches: Enhancing the transparency and interpretability of AI-powered security solutions, particularly in anomaly detection. iii) An exploration of emerging trends in Generative AI (Gen-AI) and NLP: Examining their potential to simulate and mitigate cyber threats through advanced techniques like threat intelligence generation and attack simulations. iv)Application of GenAI in cybersecurity and real-world products of GenAI for cyber security. This research aims to advance the state-of-the-art in AI-driven cybersecurity by providing insights into effective and reliable solutions for mitigating cyber risks and improving the overall security posture.
- 3 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 44662-44706
Top Cited PapersOpen AccessCyber Attack Prediction: From Traditional Machine Learning to Generative Artificial Intelligence
The escalating sophistication of cyber threats poses significant risks to individuals, organizations, and nations. Cybercrime, encompassing activities like hacking and data breaches, has severe economic and societal consequences. In today’s interconnected world, robust cybersecurity measures are paramount to mitigate these risks and protect sensitive information. However, traditional security solutions struggle to keep pace with the evolving threat landscape. Artificial Intelligence (AI) offers a powerful arsenal of techniques to address these challenges. This paper explores the application of AI methods, including Machine Learning (ML), Deep Learning (DL), Natural Language Processing (NLP), Explainable AI (XAI), and Generative AI, in solving various cybersecurity problems. This paper presents a comprehensive analysis of AI techniques for enhancing cybersecurity. Key contributions include: i) A comparative study of ML and DL methods: Evaluating their accuracy, applicability, and suitability for various cybersecurity challenges. ii) An investigation into XAI approaches: Enhancing the transparency and interpretability of AI-powered security solutions, particularly in anomaly detection. iii) An exploration of emerging trends in Generative AI (Gen-AI) and NLP: Examining their potential to simulate and mitigate cyber threats through advanced techniques like threat intelligence generation and attack simulations. iv)Application of GenAI in cybersecurity and real-world products of GenAI for cyber security. This research aims to advance the state-of-the-art in AI-driven cybersecurity by providing insights into effective and reliable solutions for mitigating cyber risks and improving the overall security posture.