RAGHU DHUMPATI

Dr. RAGHU DHUMPATI

Department of Computer Science and Engineering, Bahrain Polytechnic, Madīnat ‘Īsá, Al Janūbīyah, Bahrain
orcid
WHERE THERE IS A WILL THERE IS A WAY

Research Focus & Metrics

Academic Career Length

5Years (2022-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
27 September 2025
Verified
Conference Paper
Closed access
RDRAGHU DHUMPATI
RAGHU DHUMPATI
LKLakshmi Chandrakanth Kasireddy
Lakshmi Chandrakanth Kasireddy
GVGunavardini V
Gunavardini V
SSSanjay Kumar Suman
Sanjay Kumar Suman
LBL. Bhagyalakshmi
L. Bhagyalakshmi
STSuresh Talwar
Suresh Talwar
Edge computing represents a transforming technology that speeds up applications through data processing proximal to origins thus reducing both transmission times and network utilization. As a vital enhancement of edge computing frameworks machine learning (ML) demonstrates its capability to optimize resource allocation performance and task offloading processes and data analytical efficiency. This document describes a brand-new framework that integrates ML within edge computing to serve real-time applications while managing their resource limitations and performance delays. The proposed framework uses adjustable and compact ML models consisting of supervised learning methods with unsupervised learning and reinforcement learning to estimate workloads and ensure time-sensitive task prioritization along with resource management. Our system achieves a 30% decrease in average latency performance due to edge-based intelligence placement which produces better results than regular edge computing approaches. The system architecture operates through three interconnected layers starting from the device level to the edge and cloud level which optimizes local data processing to decrease remote cloud server data transfers. Contemporary experimental studies indicate major progress in three essential domains namely IoT, healthcare facilities and autonomous driving systems through reduced delays combined with better resources usage and increased business task success rates. The research contributes additional strength to edge intelligence research through its scalable real-time data processing method which enables smart cities and next-generation communication network development.

15 October 2025
Verified
Conference Paper
Closed access
BVB. Vasantha
B. Vasantha
RDRAGHU DHUMPATI
RAGHU DHUMPATI
The agricultural sector faces significant challenges in sustaining crop yield due to the increasing prevalence of plant diseases. Recent advancements in deep learning (DL) have introduced scalable and accurate methods for image-based crop disease diagnosis. However, despite notable progress, limitations persist in model generalization, interpretability, and comparative assessment with classical approaches. This survey critically examines state-of-the-art DL frameworks, including EfficientNet variants, attention mechanisms like Convolutional Block Attention Module (CBAM), and advanced augmentation strategies such as CutMix and CLAHE. We explore the integration of Explainable AI (XAI) techniques—Grad-CAM++, SHAP, and LIME—into plant disease detection models to enhance interpretability for agronomists. A taxonomy of datasets, including GPID-22, PlantSeg, and real-world noisy benchmarks like the Rice Leaf Disease dataset, is presented. Furthermore, we compare the performance and robustness of DL architectures with traditional machine learning classifiers such as SVM, KNN, and Random Forest using hybrid feature embeddings. The survey highlights the critical need for explainability, benchmarking, and uncertainty quantification in agricultural AI applications. By consolidating recent findings and identifying persistent research gaps, this work aims to guide future developments toward trust-worthy, transparent, and field-deployable crop disease prediction systems.

2 July 2025 in Scientific Reports
Verified
Journal Article
Open access
PDF file
RDRAGHU DHUMPATI
RAGHU DHUMPATI
ASArchana Sasi
Archana Sasi
SBShaik Johny Basha
Shaik Johny Basha
SVSuryanarayana Vadhri
Suryanarayana Vadhri
PSP. Saroja
P. Saroja
RVRamesh Vatambeti
Ramesh Vatambeti
Social media has become an integral part of daily life, with platforms like Twitter serving as popular outlets for users to share information and express grievances. While social media offers numerous benefits, it can also be misused for cyberbullying, such as insults and harassment. This highlights the importance of detecting and mitigating cyberbullying to ensure a safe online environment. Detecting sarcasm in social media posts is challenging and requires advanced automated systems. To address this, a deep learning-based approach was developed using a Convolutional Neural Network (CNN) for analysis and an Attention Mechanism-based Bidirectional Long Short-Term Memory with Gated Recurrent Unit (AM-BLSTM-GRU) for prediction. Sarcasm detection datasets were gathered from sources like Kaggle and News Headlines. Standard NLP-based auxiliary features were extracted, and an embedded CNN model refined these features into feature vectors. The AM-BLSTM-GRU model then performed sarcasm detection and sentiment classification. The Enhanced Sinogramic Red Deer (ESRD) optimizer was utilized to optimize the classifier parameters effectively. The proposed model outperformed existing deep learning approaches on popular benchmark datasets and evaluation metrics, demonstrating its effectiveness in detecting cyberbullying. This approach was validated using well-known datasets and metrics, confirming its efficacy in identifying and reducing harmful online behaviors.

1 January 2024 in International Journal of Advanced Computer Science and Applications
Verified
Journal Article
Open access
PDF file
RDRAGHU DHUMPATI
RAGHU DHUMPATI
HTHrudaya Kumar Tripathy
Hrudaya Kumar Tripathy
RBRaiza Borreo
Raiza Borreo
The constant need for robust and efficient COVID-19 detection methodologies has prompted the exploration of advanced techniques in medical imaging analysis. This paper presents a novel framework that leverages Graph Convolutional Neural Networks (GCNNs) to enhance the detection of COVID-19 from CT scan and X-Ray images. Hence, the GCNN parameters were tuned by the hybrid optimization to gain a more exact detection. Therefore, the novel technique known as Hybrid NADAM Graph Neural Prediction (NAGNP). The framework is designed to achieve efficiency through a hybrid optimization strategy. The methodology involves constructing graph representations from Chest X-ray or CT scan images, where nodes encapsulate critical image patches or regions of interest. These graphs are fed into GCNN architectures tailored for graph-based data, facilitating intricate feature extraction and information aggregation. A hybrid optimization approach is employed to optimize the model's performance, encompassing fine-tuning of GCNN hyperparameters and strategic model optimization techniques. Through rigorous evaluation and validation using diverse datasets, our framework demonstrates promising results in accurate and efficient COVID-19 diagnosis. Integrating GCNNs and hybrid optimization presents a viable pathway toward reliable and practical diagnostic tools in combating the ongoing pandemic.

18 February 2025
Verified
Conference Paper
Closed access
BKB. Kiranmayee
B. Kiranmayee
MDM. Sharmila Devi
M. Sharmila Devi
KSKalva Susheela
Kalva Susheela
RDRAGHU DHUMPATI
RAGHU DHUMPATI
KPKiran Kumar Reddy Penubaka
Kiran Kumar Reddy Penubaka
UNU Ganesh Naidu
U Ganesh Naidu
This paper proposes an integrated approach to build up IDS with proper effectiveness toward the rising need for strong network security. Network traffic anomaly detection and classification are one of the major aims and enhance the security layer against various types of cyber threats. This study is a methodical approach in which a diverse set of data is first extracted from Kaggle. The collected dataset is a comprehensive one that includes various kinds of network traffic data. The first step includes preprocessing the data, i.e., handling missing values, removing erroneous entries, and dealing with outliers using the Interquartile Range (IQR) method. To counter class imbalance, Synthetic Minority Oversampling Technique (SMOTE) is utilized in generating synthetic samples in underrepresented classes for generalization of models. The feature selection algorithm relies on the Chi-Square Lasso Feature Selection method. This method unites Chi-Square tests to select statistically significant features with Lasso regularization for feature dimension reduction with the goal of overfitting minimization. For building models, a hybrid of Support Vector Machine and K-Nearest Neighbors, which is termed as SVNN, is used. By using soft voting to aggregate the outputs of SVM and KNN, the proposed SVNN model improves the prediction accuracy, the degree of robustness, and adaptability. The presented methodology has achieved high classification accuracy 93.28%, precision 93.49%, recall 93.28%, F1 score 93.23%, and low RMSE of 0.2592. These metrics demonstrate that the model is reliable and of practical use for intrusion detection in dynamic, high-dimensional environments of networks, providing a proper solution to modern security network challenges.

4 June 2024
Verified
Journal Article
Open access
PDF file
RDRAGHU DHUMPATI
RAGHU DHUMPATI
RBR. Benschwartz
R. Benschwartz
GAGowtham A
Gowtham A
JAJeeva A
Jeeva A
SDS. Devikala
S. Devikala
KSK. Sivakumar
K. Sivakumar
This research aims to provide an analysis of the fuzzy control applications in the management of Brushless DC motors for low-cost two-wheelers. Specifically, it is possible to speak about the speed control, torque control, and the optimization of the efficiency of motors using fuzzy logic algorithms that are designed to meet the needs and requirements of the final applications of two-wheelers. To conduct trials, ten tests were selected. The results show that the fuzzy control system successfully controlled motor speed. The mean absolute error ranged from 4.6 to 5.5 rpm, and the root mean square error ranged from 6.5 to 8.2 rpm. Torque control was precise and stable with torque ripple between 0.13 and 0.16 Nm and control tracking error of 4.6 to 5.5%. Thus, the performance and efficiency of BLDC motors for low-cost electric two-wheelers can be optimized. The determined energy conversion efficiency values range from 92.3% to 93.2%. The power losses vary within from 110 to 127 watts, and the given thermal efficiency ranges from 84.5% to 86.5%. In such a way, a fuzzy logic approach is found to be a viable solution to improve the performance and efficiency of BLDC motors for low-cost electric two-wheelers and develop sustainable means of transportation.