TPThanapal Pandi

Thanapal Pandi

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Thanapal Pandi has a total of 39 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    A hybrid graph attention network and proximal policy optimization framework for adaptive fog-cloud offloading and frameproof-based secure IoT data transmission

    by Devendran Sudalai, Thanapal Pandi

    Introduction: Internet of Things (IoT) applications require secure, low-latency data transmission in resource-constrained environments. Conventional cryptographic security is computationally intensive, and static scheduling adapts poorly to changing IoT-Fog-Cloud conditions. Methods: A hybrid Graph Attention Network-Proximal Policy Optimization (GAT-PPO) framework is proposed. The IoT-Fog-Cloud infrastructure is modeled as a dynamic graph, from which GAT learns topology-aware node embeddings. A PPO agent uses these embeddings, together with task urgency, battery level, resource availability and task sensitivity, to select fog or cloud nodes for offloading. For high-sensitivity tasks, a lightweight frameproof combinatorics-based verification mechanism integrated with SHA-256 hashing is activated, and its outcome is incorporated into the PPO reward. The framework was implemented in Python and evaluated in a simulated fog-cloud environment using tasks derived from TON_IoT network traffic, over 10 independent runs. Results: The framework achieved an average latency of 68.2 ± 0.7 ms, energy consumption of 31.5 ± 0.4 mJ/task, throughput of 912 ± 9.8 tasks/s, packet delivery ratio of 98.7 ± 0.3% and security success rate of 98.9 ± 0.2%, outperforming the Without GAT, GCN-PPO and GAT-DQN configurations (Welch's t-test, p < 0.001). Under representative attacks, the false acceptance rate ranged from 0.10% to 0.60% and the false rejection rate was 0.20%. Discussion: The framework provides an effective approach for adaptive and secure IoT task offloading under the evaluated simulation conditions and shows potential for latency-sensitive IoT applications. Hardware-based validation and broader deployment studies remain future work.

  • Open Access

    Federated deep learning for stroke detection: a privacy-preserving, explainable CNN-LSTM framework

    by Reeta Samuel, Thanapal Pandi

    Strict privacy regulations in healthcare limit the centralized sharing of medical imaging data, creating challenges for developing robust and generalizable diagnostic models. Federated learning (FL) offers a privacy-preserving paradigm in which participating institutions collaboratively train a global model without directly exchanging patient images. This study presents a federated CNN-LSTM framework for privacy-preserving brain stroke classification from MRI scans, designed to learn discriminative spatial and sequential representations while retaining training data at the local client level. The proposed framework employs a CNN for hierarchical spatial feature extraction and an LSTM to model sequential dependencies among extracted features, followed by federated averaging (FedAvg) to construct the global model from local updates. Experimental evaluation on a private MRI dataset comprising 3,125 scans distributed across three non-IID clients enabled assessment under heterogeneous data distributions. The proposed federated framework achieved an accuracy of 95.20%, compared to 95.95% for centralized training, demonstrating that privacy-preserving decentralized training can achieve competitive diagnostic performance with only a marginal reduction in accuracy. Furthermore, Grad-CAM was employed to provide visual explanations of the model’s predictions by highlighting the image regions that contribute to stroke classification. In five-fold cross-validation, the Federated CNN-LSTM model achieved an average accuracy of 92.10% ± 1.50%, indicating consistent performance across folds. The findings demonstrate the feasibility of combining federated learning, hybrid CNN-LSTM representation learning, and model interpretability for privacy-preserving MRI-based stroke classification, while highlighting the need for multi-center validation and larger federated deployments before clinical translation.


  • Open Access

    MRI-based brain stroke classification using a hybrid vision transformer–BiLSTM architecture

    by Reeta Samuel, Thanapal Pandi

    Introduction: Stroke is a prominent cause of long-term disability, impacting patients’ socioeconomic status in daily life. Hemorrhagic and ischemic strokes differ in dimensions, forms, and locations, posing challenges for automated detection. Magnetic resonance imaging (MRI), particularly diffusion-weighted imaging (DWI), reveals changes in fluid balance, thereby enabling early detection. Hence, MRI scans are more accurate than computed tomography (CT) scans due to their increased sensitivity. Methods: To categorize brain strokes, a hybrid model combining bidirectional long short-term memory (BiLSTM) with a vision transformer (ViT) was developed using an MRI dataset from a private source. ViT identifies qualities using MRI. The ViT captures global contextual and spatial representations using patch-based self-attention (16×16 patches, 256-dimensional projections, four transformer encoder layers with eight attention heads), whereas the BiLSTM network (128 and 64 units) models dependencies inside transformer-encoded features. A comparative study was conducted for the hybrid architecture with deep learning models, including a convolutional neural network (baseline, 85.5%), VGG16 (87.8%), ResNet50 (89.2%), ViT (91.3%), and BiLSTM (88.6%). Results: The hybrid ViT–Bi-LSTM model achieved a precision of 97.35%, recall of 93.04%, accuracy of 95.21%, F1-score of 95.15%, and ROC-AUC of 99.36%, outperforming other comparative approaches. The standalone ViT achieved an accuracy of 91.3%, exceeding the CNN-based methods. In 5-fold cross-validation, the hybrid ViT-BiLSTM model achieved an average accuracy of 96.61%, with a standard deviation of 0.78, indicating stable performance across folds. These findings validate the combination of bidirectional temporal modeling with transformer-based feature extraction. Conclusion: By capturing the global spatial context through self-attention and bi-directional features via recurrent processing, ViT with Bi-LSTM networks expands stroke classification from MRI data. The ViT–Bi-LSTM model showed a promising approach for clinical decision support systems in early stroke diagnosis. Future research will use federated learning (FL) to protect privacy and assess model generalizability across multi-institutional MRI datasets.


  • Open Access

    Optimizing brain stroke detection with a weighted voting ensemble machine learning model

    by Reeta Samuel, Thanapal Pandi

    Brain stroke is a medical trauma that occurs when there is an impairment or decrease in blood circulation to a particular part of the brain, causing adjacent brain cells to die. Stroke diagnosis after an event is an ineffective method; other more labour-intensive and costly procedures exist for stroke diagnosis. This method involves directing a machine learning algorithm to a marked dataset to identify samples and irregularities indicative of stroke occurrence. This study focused on developing an ensemble machine learning model to predict brain stroke. The model combined the predictions of multiple individualistic classifiers, including random forest, eXtreme gradient boosting, and histogram-based gradient boosting, to improve accuracy. The proposed weighted voting-based ensemble (WVE) classifier model achieved an accuracy of 92.31% on a private stroke prediction dataset. The pre-assessment of stroke risk diagnosis, as suggested in this study, enables many people to take preventive actions well in advance, thereby lowering fatal effects. Our proposed method presents a feasible option for the early or initial diagnosis of stroke, as traditional methods, such as computed tomography (CT) scans and magnetic resonance imaging (MRIs), are time-consuming and costly. Future research could explore the use of intelligence-based optimization to enhance classification accuracy and address this limitation.


  • Open Access

    RGB Pixel Combination Method for Underwater Camouflaged Image Categorization Using Graph Neural Networks

    by K Dhamodharan, Thanapal Pandi

    Underwater imaging applications require sophisticated image processing techniques aided by Computer Vision (CV). The intensity and light variations under the water demand camouflaged object, plant, and living organism detection in deep seas. This article introduces the Edge-Recognition Method using Graph Neural Networks (ERM-GNN) to address visual differentiations of such particles with precise categorization. This method is reliable in detecting camouflaged object edges and classifying their foregrounds and backgrounds. In the edge detection and object classification process, color features are utilized, relying on RGB variants observed near the edges of the foreground and background images. The edge depth intensity feature is trained repeatedly using the GNN, forming layers between the initial and final pixels of the regions. This approach imposes an xCy pixel combination to analyze depth intensity until the foreground and backgrounds are categorized. The neural network trains the hidden layers by differentiating categorized and non-categorized RGB variants for in-depth detection. Consequently, this combination increases the chances of camouflaged underwater object categorization with better precision. This method outperforms state-of-the-art approaches such as G+L-CET, FEGM, and MFEM by a wide margin, improving recognition accuracy by 14.48%, precision by 12.81%, and object classification by 9.82%. These noteworthy improvements highlight the method’s efficacy and potential use in marine biology and environmental surveillance, necessitating accurate underwater object detection.

  • Intelligent Transportation Systems

    by S. Devendran, P. Thanapal

    The rise of IoT devices and big data generation pose challenges to Intelligent Transportation Systems (ITS), particularly in remote areas. Unmanned aerial vehicles (UAVs) have shown potential in enhancing computation and communication, but there is a lack of literature on a well-developed model that combines UAVs with a multi-hop, fog-cloud collaborative architecture. This work proposes an optimized multi-hop fog-cloud offloading architecture that uses UAVs to improve job offloading and resource allocation in ITS. The approach combines fog and cloud computing with UAV-enabled multi-hop collaboration to address the limitations of fog computing devices in processing large amounts of data and time-sensitive tasks. The offloading problem is formulated as a mixed integer nonlinear programming (MINLP) problem with constraints on user association, UAV positioning, task offloading, and resource allocation. The proposed multi-hop collaborative algorithm minimizes energy consumption, latency, and optimizes job prioritization across fog and cloud layers.

  • Top Cited Papers
    Open Access

    EMO-TS: An Enhanced Multi-Objective Optimization Algorithm for Energy-Efficient Task Scheduling in Cloud Data Centers

    by S. Nambi, P. Thanapal

    The rapid expansion of cloud data centers, driven by the increasing demand for diverse user services, has escalated energy consumption and greenhouse gas emissions, posed severe environmental risks, and increased operational costs. Addressing these challenges requires innovative solutions for optimizing resource allocation without compromising service quality. This paper presents the Enhanced Multi-Objective Optimization Algorithm for Task Scheduling (EMO-TS). This novel approach integrates Deep Reinforcement Learning (DRL) and Enhanced Electric Fish Optimization (EEFO) to create an adaptive, dynamic, and energy-efficient scheduling framework. The primary objective of EMO-TS is to significantly reduce the energy consumption of cloud data centers while maintaining high levels of resource utilization, time efficiency, and service quality. Through the hybrid methodology of DRL and EEFO, EMO-TS dynamically adjusts task scheduling based on real-time workloads and operational conditions, effectively minimizing power consumption without sacrificing system performance. Additionally, EMO-TS introduces improvements in makespan and task execution, ensuring timely completion and optimal resource use. A comprehensive set of experiments and simulations demonstrates the practical implications of EMO-TS’s results. EMO-TS outperforms traditional scheduling approaches, reducing energy consumption by up to 25% and decreasing makespan by 15%. These results underscore the algorithm’s potential to address cloud service providers’ economic and environmental concerns, offering a practical solution for green cloud computing initiatives. Furthermore, the integration of renewable energy sources within the EMO-TS framework shows potential for further reducing the carbon footprint of cloud operations, aligning with global sustainability goals.


  • Open Access

    IoT-Enabled Chlorine Level Assessment and Prediction in Water Monitoring System Using Machine Learning

    by Chandru Vignesh Chinnappan, Alfred Daniel John William, Surya Kalyan Chakravarthy Nidamanuri, S. Jayalakshmi, Ramadevi Bogani, P. Thanapal, Shahada Syed, Boppudi Venkateswarlu, Jafar Ali Ibrahim Syed Masood

    The significance of user participation in sustaining drinking water quality and assessing other factors, such as cleanliness, sanitary conditions, preservation, and waste treatment, is essential for preserving groundwater quality. Inadequate water quality spreads disease, causes mortality, and hinders socioeconomic growth. In addition, disinfectants such as chlorine and fluoride are used to remove pathogens, or disease-causing compounds, from water. After a substantial amount of chlorine has been added to water, its residue causes an issue. Since the proposed methodology is intended to offer a steady supply of drinkable water, its chlorine concentration must be checked in real-time. The suggested model continually updates the sensor hub regarding chlorine concentration measurements. In addition, these data are transmitted over a communication system for data analysis to analyze chlorine levels within the drinking water and residual chlorine percentage over time using a fuzzy set specifically using a decision tree algorithm. Additionally, a performance investigation of the proposed framework is undertaken to determine the efficiency of the existing model for predicting the quantity of chlorine substance employing metrics such as recall, accuracy, F-score, and ROC. Henceforth, the proposed model has substantially better precision than the existing techniques.


  • Open Access

    Correction to: Terahertz Micro-Strip Patch Antenna Design and Modelling for 6G Mobile Communication

    by P. Jeyakumar, J. Anandpushparaj, P. Thanapal, S. Meenatchi, M. Dhamodaran