Authors: Sai Rithwik Nooguri

Abstract: The emergence of Large Language Models (LLMs) represents one of the most consequential shifts in the history of artificial intelligence (AI) and natural language processing (NLP). Built on the Transformer architecture with self-attention mechanisms, LLMs such as BERT, GPT-3, T5, LLaMA, and GPT-4 have achieved state-of-the-art performance across a broad spectrum of linguistic tasks, fundamentally reshaping how machines comprehend and generate human language. This survey presents a systematic and comprehensive review of the evolution of NLP—from rule-based and statistical methods to the current era of foundation models—examining key architectural innovations, pre-training objectives, fine-tuning strategies including parameter-efficient methods such as Low-Rank Adaptation (LoRA), and alignment techniques including Reinforcement Learning from Human Feedback (RLHF). We critically assess performance across standard benchmarks including GLUE, SuperGLUE, and MMLU, and analyze persistent challenges such as hallucination, bias, computational cost, and explainability. Furthermore, we explore the expanding landscape of LLM applications in healthcare, education, legal reasoning, and code generation, and outline promising future directions including multimodal models, efficient inference, and AI alignment. This work aims to serve as both an accessible introduction and a scholarly reference for researchers and practitioners engaged with the rapidly evolving frontier of AI-powered language understanding.

DOI: https://doi.org/10.5281/zenodo.19630886

Published by: vikaspatanker

Authors: Shruti Chouhan, Prof. Pankaj Raghuwanshi

Abstract: Skin cancer is one of the most prevalent and rapidly increasing forms of cancer worldwide, making early detection essential for improving patient survival and treatment outcomes. Traditional diagnostic methods rely heavily on visual examination and dermoscopic analysis by dermatologists, which may sometimes be subjective and dependent on clinical expertise. In recent years, machine learning (ML) and deep learning (DL) techniques have emerged as powerful tools for automated skin cancer detection and classification. These techniques utilize medical image datasets, particularly dermoscopic images, to identify patterns and features associated with malignant and benign skin lesions. This review presents a comprehensive analysis of recent research on ML and DL-based approaches for automated skin cancer detection. Various algorithms such as Support Vector Machines (SVM), Random Forest, Convolutional Neural Networks (CNN), and transfer learning models are examined in terms of their methodologies, datasets, and performance metrics. Additionally, this study highlights the advantages, limitations, and challenges associated with these techniques. The review also discusses future research directions, including the development of more diverse datasets, interpretable models, and integration of AI-based systems into clinical practice to enhance diagnostic accuracy and healthcare efficiency.

Published by: vikaspatanker

Authors: Anushka Rastogi, Priya Gupta

Abstract: Urban areas are rapidly expanding and this brings with a set of complicated problems. Things like traffic jams, rising energy bills, overflowing landfills, and public safety concerns are becoming everyday issues for city residents. This paper looks at how artificial intelligence can play a practical role in fixing these problems. AI is making everything rapid and faster, from managing road signals to making everything work wisely and securely. At the same time, the paper does not ignore the hurdles, like data security, the cost of setting up these systems, and making sure that benefits reach every part of society, not just wealthy neighbourhoods. This study also examines how artificial intelligence can help cities better prepare for long-term challenges like climate change, population growth, and natural disasters. By looking at existing AI projects around the world, this research aims to give a realistic view of the current state and potential of AI in city infrastructure.

DOI: https://doi.org/10.5281/zenodo.19630016

Published by: vikaspatanker

Authors: Suraj Yadav, Sagar Gupta, Tanishq Raj Mahaur, Mr. Anurag Anand Duvey

Abstract: The conventional electronic voting machines often have problems like, centralized vulnerabilities, lack of transparency, prone to single-point-of-failure attacks, as well as high administrative overhead for voter verification. This paper presents a solution: A Decentralized e-voting framework created on the Ethereum Virtual Machine (EVM) that tackles these issues through the integration of blockchain immutability and mobile-native biometric authentication. With the medium this project, we propose a system that implements Solidity smart contracts to manage election lifecycles and a React Native frontend for User friendly UI and cross-platform accessibility. The key innovations such as a cycle-based state management mechanism for optimizing the contract reusability and a cryptographic credential-hashing protocol that safeguards voter identity without the need for high-cost third-party verification services.

Published by: vikaspatanker

Authors: Thakur Bhargavi Walmik, Khatate Manasvi Santosh, Prof. K. R. Metha

Abstract: The rapid growth of internet users has led to an increase in phishing attacks, where attackers create deceptive URLs to steal sensitive information. This study presents an ensemble machine learning framework for detecting phishing websites using Natural Language Processing (NLP) and multiple classifiers, including Artificial Neural Networks (ANN), Naïve Bayes (NB), Random Forest (RF), and Support Vector Machines (SVM). By extracting key features from URLs and applying machine learning techniques, the proposed model enhances detection accuracy. Comparative analysis demonstrates its effectiveness, achieving 98.4% accuracy in distinguishing phishing sites from legitimate ones. This approach offers a proactive solution to mitigate online security threats and protect users from cyber fraud. Phishing attacks have become more sophisticated, using deceptive URLs to target unsuspecting users. This research introduces a hybrid machine learning-based detection model that enhances accuracy through an ensemble of classifiers. The system utilizes Natural Language Processing (NLP) to extract critical URL features, which are then analyzed using Artificial Neural Networks (ANN), Naïve Bayes (NB), Random Forest (RF), and Support Vector Machines (SVM). Machine learning techniques are particularly effective in detecting zero-hour phishing attacks and adapting to emerging threats. Our implementation achieved a 98.4% accuracy in classifying websites as phishing or legitimate.

DOI: https://doi.org/10.5281/zenodo.19626172

Published by: vikaspatanker

Authors: Dr. A.P Srivastava, Sanjivani sharma, Mayank Kumar Singh, Aman, Saurabh Yadav, Akhand Pratap Vishwakarma

Abstract: Brain tumors are among the most critical neurological disorders and require early and accurate diagnosis to improve patient survival rates. Traditional methods of tumor detection rely heavily on manual analysis of medical images such as Magnetic Resonance Imaging (MRI), which can be time-consuming and prone to human error. This study presents a machine learning–based approach for the automated detection and classification of brain tumors from MRI images. The proposed system utilizes image preprocessing techniques to enhance image quality and remove noise, followed by feature extraction to identify significant patterns associated with tumor regions. Various machine learning algorithms, such as Support Vector Machines (SVM), Random Forest, and Convolutional Neural Networks (CNN), are applied to classify MRI images into tumor and non-tumor categories, and further categorize tumor types. The model is trained and evaluated on a labeled MRI dataset to ensure accuracy and reliability. Experimental results demonstrate that the proposed method improves diagnostic efficiency and achieves high classification accuracy compared to traditional approaches. This automated system can assist radiologists and healthcare professionals in early tumor detection, reducing diagnosis time and improving treatment planning.

DOI: https://doi.org/10.5281/zenodo.19626168

Published by: vikaspatanker

Authors: Dr. Pawan Bhaladhare, Vaibhav Ingle, Sakshi Phatake, Rushi Jagtap

Abstract: The rapid growth of internet users has led to an increase in phishing attacks, where attackers create deceptive URLs to steal sensitive information. This study presents an ensemble machine learning framework for detecting phishing websites using Natural Language Processing (NLP) and multiple classifiers, including Artificial Neural Networks (ANN), Naïve Bayes (NB), Random Forest (RF), and Support Vector Machines (SVM). By extracting key features from URLs and applying machine learning techniques, the proposed model enhances detection accuracy. Comparative analysis demonstrates its effectiveness, achieving 98.4% accuracy in distinguishing phishing sites from legitimate ones. This approach offers a proactive solution to mitigate online security threats and protect users from cyber fraud. Phishing attacks have become more sophisticated, using deceptive URLs to target unsuspecting users. This research introduces a hybrid machine learning-based detection model that enhances accuracy through an ensemble of classifiers. The system utilizes Natural Language Processing (NLP) to extract critical URL features, which are then analyzed using Artificial Neural Networks (ANN), Naïve Bayes (NB), Random Forest (RF), and Support Vector Machines (SVM). Machine learning techniques are particularly effective in detecting zero-hour phishing attacks and adapting to emerging threats. Our implementation achieved a 98.4% accuracy in classifying websites as phishing or legitimate.

Published by: vikaspatanker

Authors: Sachin Kalmani, Pratham Shinde

Abstract: Sentiment analysis has become an essential technique for extracting actionable insights from user-generated content on e-commerce platforms. This study presents a comparative analysis of machine learning and lexicon-based approaches for sentiment classification of Amazon India product reviews, where sentiment labels are derived from user star ratings. Three machine learning models — Naive Bayes, Logistic Regression, and Random Forest — are evaluated alongside two lexicon-based methods, VADER and TextBlob. Text preprocessing and feature extraction are performed using standard natural language processing (NLP) techniques combined with TF-IDF vectorization. Models are tested under four train-test split configurations (80-20, 60-40, 40-60, and 20-80) to systematically assess the effect of training data size on performance. Results show that machine learning models consistently outperform lexicon-based approaches across all evaluation metrics. At the 80-20 split, Random Forest achieves the highest accuracy of 96.22%, followed by Logistic Regression at 85.97% and Naive Bayes at 80.58%. Lexicon-based methods plateau near 73-74% accuracy across all split configurations, confirming their insensitivity to training data volume. A notable finding is that at reduced training sizes, Naive Bayes (69.65% at 20-80) underperforms both VADER (74.11%) and TextBlob (72.90%), suggesting that lexicon-based methods are more reliable when labelled training data is scarce. These findings offer practical guidance for model selection in real-world sentiment analysis applications.

DOI: https://doi.org/10.5281/zenodo.19625686

Published by: vikaspatanker

Authors: Rahul Suthar, Navneet Kumar Singh

Abstract: Instagram is widely used by handicraft sellers be- cause it offers a visual showcase for their products. However, Instagram lacks a proper product catalogue and easy inventory tracking. This paper proposes InstaCraft, a simple web CMS built with the MERN stack (MongoDB, Express.js, React.js, Node.js). InstaCraft provides a responsive product catalogue, a basic admin interface for CRUD operations on products, and direct messaging links (WhatsApp click-to-chat and Instagram profile links). The frontend is designed for mobile devices, and images are optimised (resized, converted to WebP, lazy-loaded) to reduce load times. The paper includes TikZ diagrams of the system architecture and data flow, along with example REST API and schema code. Preliminary testing indicates that InstaCraft pages load approximately 30% faster than a raw Instagram feed displaying similar content. Feature tables compare In- staCraft, Instagram-only selling, and full e-commerce platforms. InstaCraft offers a practical middle ground: it requires less effort than a full online store while providing more organisation than Instagram alone.

Published by: vikaspatanker

Authors: Sree Vishal G, Sarika K, Dr. K. Geetha

Abstract: The rapid advancement of digital technologies and the widespread use of social media platforms have significantly influenced the way individuals present themselves in modern society. Personal appearance, grooming, and aesthetic presentation have become essential aspects of self-expression and identity. However, selecting appropriate styles, outfits, and visual themes that align with individual preferences and current trends remains a complex and time-consuming task. Traditional approaches rely heavily on manual browsing, personal judgment, and external opinions, which often lack accuracy, consistency, and personalization.To address these challenges, this paper presents Aesthetiq, an Artificial Intelligence-based aesthetic analysis and personalized styling recommendation system. The proposed system is designed as a web-based application that leverages machine learning techniques to analyze user inputs such as facial images, style preferences, and visual attributes. The system performs preprocessing, feature extraction, and classification to identify suitable aesthetic categories and generate personalized recommendations.The architecture of the system consists of a frontend interface for user interaction, a backend server for processing and communication, and an AI module for performing analysis. The database stores user data, input images, and analysis results to enable efficient retrieval and history tracking. The system ensures real-time processing and provides visually interpretable outputs through an interactive dashboard.Experimental evaluation indicates that the proposed system achieves improved accuracy and performance compared to traditional methods. The system enhances user decision-making, reduces effort, and provides tailored recommendations that align with individual preferences and modern trends.

DOI: https://doi.org/10.5281/zenodo.19614962

Published by: vikaspatanker