Category Archives: Uncategorized

Authors: Rishi Raghav Singh, Rohan Singh, Rajat Takkar

Abstract: More than 30% of worldwide deaths involve diseases caused by cardiovascular and lifestyle factors (WHO, 2023) As awareness of early risk identification advances, accessible practical screening tools for use in primary care continue to be either very expensive, reliant on specialists or both. In this paper, we propose a Multi-Modal AI-Based Health Intelligence Framework with an explicit focus on two interrelated concepts encapsulated in the form of two specialized individual modules: Disease Risk Assessment (DRA) module and Lifestyle Analysis (LSA) module. After systematic preprocessing and class-balancing, the DRA module trains LR, SVM, and RF on the Cleveland Heart Disease dataset (303 patients). The LSA module takes user-reported behavioral behaviors — BMI, physical activity, sleep, dietary quality, and stress — to calculate a composite Lifestyle Risk Index (LRI). Both modules are provided through a Streamlit web application that provides real time predictions with SHAP-based explanation. Amongst all the classifiers we evaluated, Random Forest performed best with a 91.8% accuracy, AUC-ROC = 0.956 It powers a sub 60 ms response time for the system and is deployable in the cloud.

Published by: vikaspatanker

Authors: Kasarla Vanitha, B.Archana

Abstract: Cloud computing has transformed the IT landscape by providing scalable, flexible, and cost-effective services. However, its multi-tenant infrastructure introduces significant security and privacy challenges due to shared resources and virtualized environments. This paper examines established security and privacy frameworks, threat models, and protective architectures designed to address these concerns. Through a comparative analysis of existing literature and technical frameworks, the study identifies key vulnerabilities and effective mitigation strategies in multi-tenant cloud environments. Additionally, graphical models and charts are included to demonstrate how shared resource access and virtualization can be secured using encryption, tenant isolation, and dynamic authentication mechanisms.

DOI: https://doi.org/10.5281/zenodo.19909008

Published by: vikaspatanker

Authors: Wasim Riyajoddin Kazi, Om Vitthal Devakate, Vishal Popatrao Jagadale, Kavita Shinde

Abstract: In recent years, the issues related to public safety and security have increased significantly, which resulted in a surge of demand for automated surveillance systems. However, traditional monitoring systems based on CCTV require constant human surveillance, which is not only wasteful but also error- prone. This paper proposes a deep learning-based surveillance system that can automatically detect suspicious activities in videos. The proposed model utilizes CNNs to classify video frames into normal and suspicious categories. Upon detection of suspicious activity, the system captures the frame and sends an automated email notification to the registered system administrator using the SMTP protocol. The proposed system utilizes OpenCV for video processing, TensorFlow/Keras for training and predicting the models, and SQLite to securely store administrator information within a database.

Published by: vikaspatanker

Authors: Prudvi Saisaran Ponduru

Abstract: Recent advances in foundation models, multimodal learning, reasoning-oriented large language models, agentic workflows, and edge AI have expanded the capabilities of artificial intelligence systems. However, practical decision-making remains brittle because many systems optimize prediction quality while under-modeling intervention effects, uncertainty, safety constraints, latency budgets, and human accountability. This paper introduces AEGIS-DM, an adaptive, edge-aware, governed, interventional, and safe decision-making framework designed for AI systems deployed across emerging technology settings including agentic assistants, cyber-physical systems, healthcare decision support, and enterprise automation. The framework combines five layers: multimodal state representation, predictive scoring, causal effect estimation, simulator- or planner-based long-horizon optimization, and a governance layer for calibration, fairness, policy checks, logging, and human override. We further propose a cross-domain evaluation protocol using public resources such as Adult, D4RL, WebShop, ALFWorld, MIMIC-IV Demo, NASA CMAPSS, and M5, together with open-source tooling including OpenAI Evals, Responsible AI Toolbox, OpenSpiel, RecSim NG, Stable-Baselines3, and RLlib. Because this manuscript is a methods-and-benchmark contribution, the quantitative section reports deterministic scenario-based simulation results under the stated protocol rather than production deployment measurements. Under the reference protocol, the proposed hybrid approach is expected to outperform rule-based, supervised-only, offline-RL-only, and prompt-only agent baselines in composite decision quality and robustness while maintaining substantially better latency and cost than cloud-only frontier-model pipelines.

DOI: https://doi.org/10.5281/zenodo.19968793

Published by: vikaspatanker

Authors: Rajat Srivastava, Mr. Ankit Singh, Sneha Mehrotra, Shaifali Singh, Shreyansh Srivastav

Abstract: There’s a lot more to student performance than just marks. Some kids barely pass written exams but shine in group projects or sports. The problem is, most colleges still judge students almost entirely by their test scores. That’s like judging a fish by its ability to climb a tree. By the time a teacher realizes someone’s struggling, that student might already be failing or even thinking of dropping out. So what if we could spot trouble earlier — way before the report card says it all? That’s what this paper is about. We used machine learning to sift through student data — attendance, past grades, even family background — and predict who might fall behind. Not just for the sake of prediction, but to actually give teachers a heads-up so they can step in and help. The results were pretty solid. Our model caught most at-risk students with over 90% accuracy. Not perfect, but a lot better than waiting till the end of the semester.

Published by: vikaspatanker

Authors: Avinash, Anshika, Shivam

Abstract: We know digital payment systems are growing faster all over the world, and as they grow, they have some consequences. One big problem among them is digital payment fraud, which rapidly increases as payment systems grow. Fraudsters can surpass rule-based detection systems as they adapt; they have new patterns for doing fraud. They find loopholes in the main architecture from where they manipulate data and do fraud. We study both problems and reach a very solid solution to track down all fraud patterns. We added artificial intelligence and the Hyperledger Fabric blockchain, which is used to detect the pattern of fraud, and a blockchain, which is used to make the payment system tamper-proof. All data related to the payment system are stored in a single system, which is very secure and not able to be encrypted. The detection system runs on four methods. For unstable workflow it uses LSTM networks. For rule-based classification, we used a random forest classifier. For fraud detection, we used a GraphSAGE network, and last, for any suspicious activity, we used an autoencoder. All the things are watched by a meta-learner, which analyzes and combines their output and provides data to trigger a smart contract response, which works automatically. Different financial institutions are used to train their systems without using shared row transaction data to make privacy learn their module detection. We concluded our study, but two public benchmarks are set by PaySim (6.35M transactions) and IEEE-CIS (590K transactions). In PaySim we succeed with up to 98.3% accuracy and an AUCROC of 0.991. Adversarial robustness testing shows the team requires 3.2 times larger to prevent any mistake for success for a single model. These results show much need of AI and blockchain. Using AI and blockchain is very efficient; they are better than anything else to detect fraud.

Published by: vikaspatanker

Authors: Vansh Sharma

Abstract: Banking and financial institutions are increasingly encountering the challenges of credit card fraud. Statistics suggest that each year financial institutions incur losses close to billions of dollars globally due to such frauds .Hence it is evident for financial institutions to continue to invest in advanced fraud detection systems to minimize the impact of credit card fraud on their bottom line and protect their customers from financial losses.Before deep diving into the solutions which can be proposed to solve the problem of credit card fraud , it is important to know the ways in which these frauds are taking place and what loopholes are being misused to catalyze these frauds .Hence in our research paper we first look at ways in which these frauds are taking place. Moreover, one of the other challenges to proposing a solution to this problem is the presence of highly imbalanced datasets to train the model , which motivates us to apply various techniques such as Synthetic Minority Oversampling Technique (SMOTE) to make the datasets balanced which will allow us to train the model better .We implement Artificial neural network + Recurrent neural network with auto-encoder architecture to make a model for one-class classification . The model uses these relationships to make predictions about the likelihood of fraud in new transactions. ANNs can be used to process large amounts of data and are particularly effective in detecting non-linear relationships between variables.

DOI: https://doi.org/10.5281/zenodo.19884020

Published by: vikaspatanker

Authors: Dr. Meghna Utmal, Sakshi Singh, Kunti Uikey, Vaishali Gupta, Sajal Pandey

Abstract: — Heart disease remains a major health concern worldwide, affecting a large proportion of the global population. According to reports by the World Health Organization (WHO), approximately 17.9 million deaths occur annually due to cardiovascular diseases. In the context of the COVID-19 pandemic and its post-infection complications, cardiac failure has emerged as a commonly observed condition, highlighting the critical need for early diagnosis and prediction of heart disease to enable effective prevention. Timely detection can significantly reduce mortality rates. Recent advancements in machine learning techniques have greatly contributed to the healthcare sector, particularly in the prediction of heart diseases, thereby saving numerous lives. This paper presents an efficient ensemble-based machine learning approach for predicting heart-related disorders, achieving an accuracy of 88.52%.

DOI: https://doi.org/10.5281/zenodo.19882012

Published by: vikaspatanker

Authors: Nv Subhasri, Madhunisha. A, Shruthi. T

Abstract: This dissertation examines the growing intersection of law, technology, and regulation in the context of autonomous vehicles (AVs) in India, with particular emphasis on issues of privacy, data protection, and cybersecurity. It analyzes existing Indian legal frameworks, such as the Information Technology Act and the proposed Personal Data Protection Bill, to assess whether they are equipped to handle the unique challenges posed by AV technology. The study also compares India’s approach with international standards, including the GDPR and regulatory models followed in countries like the United States and China. Through this comparative perspective, the research highlights existing gaps and suggests areas where India can strengthen its legal and regulatory response.

Published by: vikaspatanker

Authors: Rajat Takkar, Gunjan Lathwal, Devanshi Dadwal, Bhumika Aggarwal, Gaurang Batra

Abstract: AI chatbots and large language models show up almost everywhere these days – customer support, healthcare, schools, even hiring. While they’re good at handling language, there’s still a big question about bias. These systems often pick up biases from their training data and then reflect them back in their answers. That includes biases related to gender, jobs, places, or wealth. This study explores whether chatbots respond to demographic-based questions with built-in bias. We used a set of structured prompts and gathered answers from several AI chatbots, recording all responses for analysis. Every answer was examined using sentiment analysis and a neutrality scoring method. This measured how fair or unbiased each system was. We performed all our analysis using Python tools like Pandas, TextBlob, and Matplotlib. Our expectation was that chatbot responses would usually be objective, but some subtle biases could sneak in depending on how you ask the question or what the topic is. Some questions just lead to more bias than others. By scoring fairness, we can actually quantify differences and see which systems are more neutral. This approach helps assess how these AI tools deal with real-world issues and fairness.

Published by: vikaspatanker