Security on Cloud Resource Consumption of Malicious Attacks – A Systematic Review H Shaheen, S. Vinoth Kumar, S Vidhya, K Ravikumar, S Girirajan, R Senthilkumar Journal of Telecommunications and the Digital Economy, 2026 An ever-increasing number of buyers are moving to the public cloud and facilitating their applications on cloud cost-effectively to benefit from enhanced high accessibility and versatility. Malicious Cloud Bandwidth Consumption (MCBC) is one type of threat where attackers consume the cloud bandwidth slowly and continuously for an extended time, causing a financial burden to the cloud consumer. This research emphasises securing cloud web resource consumption by detailing MCBC attacks and proposes a methodology for building classifiers using Machine Learning (ML) to detect malicious requests accurately. Then, at that point, we evaluate the performance of each method based on its features, advantages, and limitations. Additionally, we also outline future research directions for developing improved cloud computing models.
Stuttering Speech Recognition System using Enhanced MFCC with Single Gated Recurrent Unit Anusha Patra, Abhignya Priyadarshini, E. Sasikala, Girirajan S Proceedings 3rd International Conference on Artificial Intelligence and Machine Learning Applications Healthcare and Internet of Things Aimla 2025, 2025 Stuttering disrupts the natural flow of speech through repetitive or prolonged sounds, syllables, or words. Traditional identification and quantification of these disfluencies by Speech-Language Pathologists are often subjective, time-consuming, and prone to human error. This research proposes an automated method for objectively detecting stuttering using enhanced Mel Frequency Cepstral Coefficients (MFCC) combined with a Single Gated Recurrent Unit (GRU) model. Utilizing the UCLASS dataset, which includes a diverse range of stuttered speech samples, we pre-process and segment the data to extract MFCC features. These features are then input into a machine learning model, specifically a Single GRU, to classify stuttering events. The model is trained and tested with careful tuning of hyperparameters to optimize its accuracy and performance in detecting stutter in speech.
Improved Detection of Kidney Stones and Fractures Using YOLO and CNN: A Comparative Study with Older Model Uday Singh Slathia, Kumar Ashish, Girirajan S, Sandhia G K, Karthikeyan M 2025 International Conference on Recent Advances in Electrical Electronics Ubiquitous Communication and Computational Intelligence Raeeucci 2025, 2025 The accurate and early detection of medical conditions like kidney stones and bone fractures is critical for effective treatment. In this study, we propose an enhanced detection approach using a combination of YOLO (You Only Look Once) and Convolutional Neural Networks (CNNs). These models are evaluated against older, traditional detection algorithms in terms of precision, recall, and computational efficiency. By utilizing YOLO’s real-time object detection capabilities and CNN’s feature extraction power, we aim to significantly improve diagnostic accuracy for both conditions. Our dataset includes medical imaging data of kidney stones and fractures, pre-processed to optimize model training and validation. A comparative analysis reveals the advantages and limitations of our method compared to the older models, highlighting improvements in detection speed, accuracy, and robustness across varying image qualities. This research demonstrates the potential of integrating YOLO and CNN for enhanced medical diagnostics and paves the way for future improvements in medical imaging technologies.
Enhanced Machine Learning Model for Heart Disease Prediction Bhavya Golchha, Hrutuja Patil, S.Girirajan, G.K.Sandhia, M.Karthikeyan 2025 Global Conference in Emerging Technology Ginotech 2025, 2025 Heart disease remains one of the leading causes of mortality worldwide, highlighting the critical need for effective prediction and early detection methods. This paper proposes a hybrid heart disease prediction model that integrates three machine learning techniques: Linear Regression (LR), Support Vector Machine (SVM), and Random Forest (RF). The hybrid approach leverages the strengths of each algorithm to improve the accuracy and robustness of predictions. Linear Regression is employed for its ability to model the relationship between risk factors and heart disease in a straightforward and interpretable way, providing baseline predictions. Support Vector Machine contributes by optimizing decision boundaries and handling non-linearities in the dataset, ensuring better classification in complex cases. Finally, Random Forest, an ensemble learning method, enhances prediction accuracy by reducing variance and avoiding overfitting, making it suitable for handling high-dimensional and noisy medical data. The proposed hybrid model is evaluated using a heart disease dataset, and its performance is compared against standalone algorithms. Key metrics such as accuracy, precision, recall, and F1-score are used to assess the effectiveness of the model. Experimental results demonstrate that the hybrid approach significantly improves prediction performance, achieving higher accuracy and lower false positive and false negative rates compared to individual models.
News2Braille: A T5-based News Summarization System for the Visually Impaired Karthikeyan M, Ananmay Anand, Jathin S. Lohi, Girirajan S, Sandhia G. K Proceedings of 6th International Conference on Intelligent Communication Technologies and Virtual Mobile Networks Icicv 2025, 2025 Access to real-time news has become increasingly seamless in the digital age; however, visually impaired individuals often face challenges in navigating conventional news platforms. This research presents a novel T5 Transformer-Based News Summarizer and Braille Translator, designed to enhance news accessibility for the visually impaired. The proposed system comprises three core components: extraction of relevant textual content from online articles using the Newspaper3k library; abstractive summarization using a Fine-tuned Text-to-Text Transfer Transformer (T5) model to retain essential information; and conversion of the summarized content into Braille, enabling tactile reading.This integrated approach empowers users to access concise news content independently, without reliance on multiple tools or applications. Future enhancements may include implementing deep learning-based Braille translation, incorporating Optical Character Recognition (OCR) for printed news, and integrating text-to-speech functionality for auditory consumption. Overall, this work contributes toward a more inclusive and accessible digital ecosystem for visually impaired individuals.
Brain Stroke Detection Using DenseNet Algorithm Saragadam Vamsi Varshith, Kotikalapudi Tulasi Venkata Bhadra Ganesh, Karthikeyan M, Girirajan S, Sandhia G K 2025 Global Conference in Emerging Technology Ginotech 2025, 2025 A brain stroke occurs when blood flow to a part of the brain is interrupted or reduced, depriving brain tissue of oxygen and nutrients. This can cause brain cells to die, leading to potential loss of function depending on the affected area. A significant amount of research has focused on using Convolutional Neural Networks (CNNs) for brain stroke detection, leveraging their ability to automatically extract features from medical images. In this paper, we will employ the DenseNet201 architecture in performing differential diagnosis from normal to stroke brain conditions using CT scans data. First and foremost, we would like to outline what is basically the key objective of developing a binary classification model that can accurately determine the presence or absence of stroke. The dataset consists of CT scans divided into two classes: normal and stroke. This model utilizes DenseNet201— which draws its power from extremely efficient use of parameters and connections—by densely feeding each other in the learning of detailed features from CT scans. Since performance varies significantly on complexity and subtle variation with fine-tuning of the model in medical images, this research is therefore significant in examining the prospects of using deep learning-based models, such as DenseNet201, to assist with stroke detection and diagnosis. Data Set is named as Brain_Stroke_CT-SCAN_image and gained an accuracy as 93%.
Signify: A Transformer-based Approach for Real-Time Speech to Sign Language Translation Kasi Viswanathan K, Karthikeyan M, Elatchuman RV, G. K. Sandhia, Girirajan S Proceedings 3rd International Conference on Self Sustainable Artificial Intelligence Systems Icssas 2025, 2025 Signify is a real-time system for translating voice to sign language, utilizing sophisticated transformer-based architectures, such as Wav2Vec 2.0 for speech recognition and Vision Transformers (ViT) for gesture representation. The system demonstrates superior performance on essential assessment criteria, with an accuracy of 96.5%, an F1-score of 96.1%, and a precision of 95.3%, surpassing traditional architectures like ConvLSTM and Transformer-TTS. The system eliminates dependence on manual transcription and enables comprehensive translation via a Flask-based web application, ensuring reliable performance in diverse acoustic environments and among different speaker profiles. Experimental assessments performed on benchmark datasets further validate the system's scalability, adaptability, and appropriateness for real-time implementation. The architecture addresses deficiencies in accessibility tools for the hearing-impaired community, providing a modular approach that allows for future enhancements, including dynamic gesture animation and language support.
Code Generation Empowered by Natural Language Processing and Machine Learning Algorithms S.Girirajan Advances in Nonlinear Variational Inequalities, 2025 The goal of this study is to revolutionize code creation processes by investigating the synergistic union of machine learning (ML) and natural language processing (NLP). Enterprising non-programmers with entrance barriers, traditional approaches to code generation frequently demand expert-level programming expertise. Development teams can communicate coding tasks in natural language by utilizing NLP techniques like language modeling and semantic parsing. This helps to close the gap between human intent and instructions that can be executed by a computer. By incorporating ML techniques, the system may also more effectively understand and produce code that is compatible with a wider range of programming languages and paradigms. This research clarifies the revolutionary potential of NLP and ML-driven code creation and highlights its consequences for software development efficiency, accessibility, and innovation through an extensive assessment of current developments and case examples.
Cybersecurity: The part played by artificial intelligence Sandhia G. K., M. Ranjani, N. Nithiyanandam, Prakash U. M., R. Srinivasan, S. Girirajan, D. Saisanthiya, J. Ramaprabha Analyzing Privacy and Security Difficulties in Social Media New Challenges and Solutions, 2024 The rapid growth of technology in recent decades has led to a sharp rise in cybercrimes, impacting millions of individuals and businesses through identity theft, data breaches, and similar threats. These cyberattacks have become more advanced and effective in their malicious intent, revealing the shortcomings of traditional cybersecurity measures. As a result, there is a growing need for innovative strategies to combat these threats, and Artificial Intelligence (AI) is emerging as a key solution. AI's flexibility, speed, and precision make it well-suited to address evolving security challenges. Companies handling sensitive data can use AI to automate the detection of threats and stay ahead of cybercriminals. AI-powered systems offer strong defense capabilities by identifying malware, detecting unusual traffic, preventing phishing and spam, and responding to data breaches and unauthorized access. This chapter outlines the significant impact of AI on cybersecurity, examines future trends, and highlights the challenges that remain.
A Hybrid Cellular Automata - Patch Based Local Principal Component Analysis Techniques for Improving Image De-Noising International Journal of Intelligent Systems and Applications in Engineering, 2024
