Venkata Nagaraju Thatha
Research Subjects
Venkata Nagaraju Thatha has a total of 23 co-authors. Here are the top co-authors by number of shared publications:
A. Sudhir Babu
PVPSIT
1 shared publication- Argula Rishitha
MLR Institute of Technology,Department of Information Technology,Hyderabad,India
1 shared publication - Budda Akanksha
MLR Institute of Technology,Department of Information Technology,Hyderabad,India
1 shared publication D. Haritha
JNTUK University
1 shared publicationD. Siri
Department of Computer Science and Engineering, Gokaraju Rangaraju Institute of Engineering and Technology, Hyderabad, India
1 shared publication- G. Uday Kiran
Department of CSE (Artificial Intelligence & Machine Learning), B.V. Raju Institute of Technology, Narsapur, Medak-502313, Greater Hyderabad, Telangana, India
1 shared publication
Latest Publications
- 4 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1500-1506
Deep Learning Powered Malicious URL Detection Using Keyword Aware Character Embeddings
The increasing complexity of malicious URLs is especially problematic for web security, as the attackers constantly use obfuscation decoding to bypass traditional black list and rule based approaches. To address this issue, a character-level deep learning-based malicious URL classification system is proposed in this paper. The proposed system adopts a hybrid CNN-GRU architecture, which exploits both local patterns of malicious URL attacks and sequential relationships in URLs by combining keyword sensitive learned character embeddings. The system is evaluated on a large dataset that encompasses benign and various forms of malicious URLs, such as phishing and directory traversal attacks [13]. Results of the experiments show that our proposed system is better than the base models with respect to classification accuracy and F1-measure in detecting obfuscated and novel malicious URL attacks. Experimental results show that the proposed model which integrates character level feature, sequence modeling and keyword sensitivity has significantly improved the detection accuracy of malicious URLs.
- 4 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1500-1506
Deep Learning Powered Malicious URL Detection Using Keyword Aware Character Embeddings
The increasing complexity of malicious URLs is especially problematic for web security, as the attackers constantly use obfuscation decoding to bypass traditional black list and rule based approaches. To address this issue, a character-level deep learning-based malicious URL classification system is proposed in this paper. The proposed system adopts a hybrid CNN-GRU architecture, which exploits both local patterns of malicious URL attacks and sequential relationships in URLs by combining keyword sensitive learned character embeddings. The system is evaluated on a large dataset that encompasses benign and various forms of malicious URLs, such as phishing and directory traversal attacks [13]. Results of the experiments show that our proposed system is better than the base models with respect to classification accuracy and F1-measure in detecting obfuscated and novel malicious URL attacks. Experimental results show that the proposed model which integrates character level feature, sequence modeling and keyword sensitivity has significantly improved the detection accuracy of malicious URLs.
- 24 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 667-672
Hybrid Machine Learning Approaches for Passive Blind Image Splicing Forensics
Digital image editing technology is moving so fast that it might completely destroy the visual content credibility. First of all, forensic specialists have trouble spotting image collisions and mixing when the edited areas are perfectly integrated and then the image is post-processed. Moreover, the passive detection using DCT, DWT, and LBP has been not very successful and, in fact, has often failed to provide even the slightest evidence that could point to a global forgery in realworld cases. In coping with these trials, the present research brings forth an innovative hybrid machine learning framework for passive blind image splicing detection that relies on the uniting of the handcrafted forensic descriptors with the deep learning-based feature representations. The technique is featured by the merging of texture, frequency, and noise-related cues, while the CNN embeddings grasp the semantic inconsistencies that splicing has introduced via fusing. The combination of these complementary features indeed enhances the capability of the system to detect tampering even after the images have undergone the aforementioned attacks like compression, rescaling, filtering, and illumination variations. The framework is tested on standard datasets such as CASIA v2.0 and the Columbia Image Splicing Dataset allowing both image-level detection and pixel-level localization. The results from the experiment show that the method outperforms both the handcrafted and deep learning approaches in terms of detection performance measures such as precision, recall, and the F1-score. The present study also points out major drawbacks of current splicing detection methods and opens up new pathways towards the trustworthiness of forensic applications in the areas of journalism, law enforcement, and social media authenticity verification.
- 24 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 667-672
Hybrid Machine Learning Approaches for Passive Blind Image Splicing Forensics
Digital image editing technology is moving so fast that it might completely destroy the visual content credibility. First of all, forensic specialists have trouble spotting image collisions and mixing when the edited areas are perfectly integrated and then the image is post-processed. Moreover, the passive detection using DCT, DWT, and LBP has been not very successful and, in fact, has often failed to provide even the slightest evidence that could point to a global forgery in realworld cases. In coping with these trials, the present research brings forth an innovative hybrid machine learning framework for passive blind image splicing detection that relies on the uniting of the handcrafted forensic descriptors with the deep learning-based feature representations. The technique is featured by the merging of texture, frequency, and noise-related cues, while the CNN embeddings grasp the semantic inconsistencies that splicing has introduced via fusing. The combination of these complementary features indeed enhances the capability of the system to detect tampering even after the images have undergone the aforementioned attacks like compression, rescaling, filtering, and illumination variations. The framework is tested on standard datasets such as CASIA v2.0 and the Columbia Image Splicing Dataset allowing both image-level detection and pixel-level localization. The results from the experiment show that the method outperforms both the handcrafted and deep learning approaches in terms of detection performance measures such as precision, recall, and the F1-score. The present study also points out major drawbacks of current splicing detection methods and opens up new pathways towards the trustworthiness of forensic applications in the areas of journalism, law enforcement, and social media authenticity verification.
- Open Access
- 21 February 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 28972-28987
Analyzing Public Sentiment on the Amazon Website: A GSK-Based Double Path Transformer Network Approach for Sentiment Analysis
Sentiment Analysis (SA) holds considerable significance in comprehending public perspectives and conducting precise opinion-based evaluations, making it a prominent theme in natural language processing research. With the increasing trend of online shopping and social media usage, there is a constant influx of diverse data types such as images, videos, audio, and text. Notably, text stands out as the most crucial form of unstructured data, demanding heightened attention from researchers. Given the voluminous nature of data, various methodologies have been proposed to effectively mine big datasets for valuable insights. The challenge of accurately identifying polarity in extensive customer evaluations persists due to the intricacies associated with handling large textual datasets derived from reviews, comments, tweets, and posts. This study addresses this challenge by presenting a straightforward architecture, the Double Path Transformer Network (DPTN), designed to model both global and local information for comprehensive review categorization. To enhance the synergy between the attention path and the convolutional path, the study advocates a parallel design that combines a robust self-attention mechanism with a convolutional network. The research employs the gaining-sharing knowledge optimization (GSK) approach to fine-tune hyperparameters, thereby improving the model’s classification accuracy. Additionally, the investigation demonstrates that optimization algorithms and deep learning collaboratively manage class imbalances with finesse, even in the absence of explicit measures for such concerns. In the experiment analysis of the proposed model ultimately achieved an accuracy of 95.
- 21 February 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 28972-28987
Open AccessAnalyzing Public Sentiment on the Amazon Website: A GSK-Based Double Path Transformer Network Approach for Sentiment Analysis
Sentiment Analysis (SA) holds considerable significance in comprehending public perspectives and conducting precise opinion-based evaluations, making it a prominent theme in natural language processing research. With the increasing trend of online shopping and social media usage, there is a constant influx of diverse data types such as images, videos, audio, and text. Notably, text stands out as the most crucial form of unstructured data, demanding heightened attention from researchers. Given the voluminous nature of data, various methodologies have been proposed to effectively mine big datasets for valuable insights. The challenge of accurately identifying polarity in extensive customer evaluations persists due to the intricacies associated with handling large textual datasets derived from reviews, comments, tweets, and posts. This study addresses this challenge by presenting a straightforward architecture, the Double Path Transformer Network (DPTN), designed to model both global and local information for comprehensive review categorization. To enhance the synergy between the attention path and the convolutional path, the study advocates a parallel design that combines a robust self-attention mechanism with a convolutional network. The research employs the gaining-sharing knowledge optimization (GSK) approach to fine-tune hyperparameters, thereby improving the model’s classification accuracy. Additionally, the investigation demonstrates that optimization algorithms and deep learning collaboratively manage class imbalances with finesse, even in the absence of explicit measures for such concerns. In the experiment analysis of the proposed model ultimately achieved an accuracy of 95.
- Open Access
- 1 October 2023
- Published by Institute of Advanced Engineering and Science in International Journal of Electrical and Computer Engineering (IJECE)
- Vol. 13 (5) , 5550-5559
Security and risk analysis in the cloud with software defined networking architecture
Cloud computing has emerged as the actual trend in business information technology service models, since it provides processing that is both cost-effective and scalable. Enterprise networks are adopting software-defined networking (SDN) for network management flexibility and lower operating costs. Information technology (IT) services for enterprises tend to use both technologies. Yet, the effects of cloud computing and software defined networking on business network security are unclear. This study addresses this crucial issue. In a business network that uses both technologies, we start by looking at security, namely distributed denial-of-service (DDoS) attack defensive methods. SDN technology may help organizations protect against DDoS assaults provided the defensive architecture is structured appropriately. To mitigate DDoS attacks, we offer a highly configurable network monitoring and flexible control framework. We present a dataset shift-resistant graphic model-based attack detection system for the new architecture. The simulation findings demonstrate that our architecture can efficiently meet the security concerns of the new network paradigm and that our attack detection system can report numerous threats using real-world network data.
- 1 October 2023
- Published by Institute of Advanced Engineering and Science in International Journal of Electrical and Computer Engineering (IJECE)
- Vol. 13 (5) , 5550-5559
Open AccessSecurity and risk analysis in the cloud with software defined networking architecture
Cloud computing has emerged as the actual trend in business information technology service models, since it provides processing that is both cost-effective and scalable. Enterprise networks are adopting software-defined networking (SDN) for network management flexibility and lower operating costs. Information technology (IT) services for enterprises tend to use both technologies. Yet, the effects of cloud computing and software defined networking on business network security are unclear. This study addresses this crucial issue. In a business network that uses both technologies, we start by looking at security, namely distributed denial-of-service (DDoS) attack defensive methods. SDN technology may help organizations protect against DDoS assaults provided the defensive architecture is structured appropriately. To mitigate DDoS attacks, we offer a highly configurable network monitoring and flexible control framework. We present a dataset shift-resistant graphic model-based attack detection system for the new architecture. The simulation findings demonstrate that our architecture can efficiently meet the security concerns of the new network paradigm and that our attack detection system can report numerous threats using real-world network data.