Category Archives: Uncategorized

Authors: Mr. Omesh Wadhwani, Bhagyashri Rahangdale, Dhanashree Dahake, Parika Pandharkar, Rishita Pokhare

Abstract: This research paper presents a detailed exploration of an AI-based traffic management system leveraging the YOLOv8n object detection model. The system aims to improve traffic flow, reduce congestion, and enhance overall road safety through real-time analysis of traffic conditions. The paper covers various aspects, including the system architecture, the implementation details of YOLOv8n for vehicle detection and tracking, the integration of detected data into a traffic management platform, and the experimental results demonstrating the system's performance and effectiveness. The study also addresses challenges in deploying AI-based traffic management systems and suggests potential solutions for future research and development. The proposed system is trained using the COCO dataset along with custom traffic video data to ensure robustness under different environmental conditions. Performance evaluation is carried out using standard metrics such as precision, recall, and detection accuracy. Experimental results show that the model achieves a precision of 0.92, recall of 0.89, and overall detection accuracy of 91%, while effectively estimating traffic density in real-time scenarios. These results demonstrate the system’s capability to support adaptive signal timing and significantly improve traffic efficiency.

DOI: https://doi.org/10.5281/zenodo.19698223

Published by: vikaspatanker

Authors: Viraja kanawally

Abstract: Artificial intelligence is becoming increasingly integrated into recruitment and is changing the paradigm in human resource management practices by helping organizations become more efficient in their hiring, decreasing time to hire, and improving quality of hire performance. The current paper explores how predictive analytics driven by AI technology can be applied within a recruitment process by automating resume screening, job-candidate match, and employee turnover predictions. Using survey data collected from 304 firms based in Europe who have adopted AI tools for recruiting purposes, it is found that AI can cut down time to hire by 48.8%, reduce cost per hire by 54.6%, and increase retention rates by 17.9%. Still, 15% of organizations adopt AI to predict internal mobility. The major reasons preventing them from doing so are fears about algorithmic bias, excessive costs associated with AI tool adoption, and resistance from applicants. A framework for predicting recruitment outcomes with the help of AI will be presented.

DOI: https://doi.org/10.5281/zenodo.19697173

Published by: vikaspatanker

Authors: Anam Siddiqui, Saima Shaikh, Sana Shaikh, Alfiya Shaikh, Prof. Nargis Shaikh

Abstract: Personal safety has become a critical concern in modern society due to the increasing rate of crimes and emergency situations. This paper presents a Human Safety Device that integrates Internet of Things (IoT) technology with dual communication systems, namely GSM-based SMS alerts and Telegram-based real-time notifications, to ensure reliable emergency response. The system is built using an ESP32 microcontroller interfaced with a GPS module for location tracking, a pulse sensor for heart rate monitoring, and an MPU6050 sensor for motion detection. In emergency conditions, triggered manually via a panic button or automatically through abnormal sensor readings, the system captures and transmits location and health data to predefined contacts. Experimental evaluation shows that the system achieves an average alert response time of 3–5 seconds for Telegram notifications and 5–10 seconds for GSM-based SMS delivery. The GPS module provides location accuracy within ±5–10 meters, while sensor readings maintain an accuracy of approximately 95% under normal conditions. Additionally, a web-based interface enables real-time monitoring and visualization of user data. The proposed system is compact, cost-effective, and highly reliable, making it suitable for real-world deployment in personal safety applications.

Published by: vikaspatanker

Authors: Manasi Waghe, Danish Chandargi, Mohammad Aamir Rayyan, Raviraj Joshi, Dr. A.R. Deshpande

Abstract: The translation of government and legal documents from Marathi to English poses unique challenges due to linguis- tic complexity, domain-specific terminology, structural richness, and low-resource constraints. General-purpose machine translation systems often fail to maintain semantic fidelity, formatting, and terminological consistency required for administrative and legal texts. This survey explores recent advances in multilingual machine translation, domain adaptation techniques, OCR-driven document understanding, Marathi-specific NLP resources, and terminology- constrained translation methods. We examine the state-of-the-art in robust Marathi-to-English translation systems and highlight critical gaps, focusing on integrating layout-aware models and domain- specific constraints to improve translation quality and reliability for official government documentation.

Published by: vikaspatanker

Authors: Srinithi G D, Hidhesh R M

 

Abstract: Pneumonia remains one of the leading causes of mortality worldwide, particularly among children under five and the elderly. Early and accurate diagnosis through chest X-ray interpretation is critical, yet manual analysis by radiologists is time-consuming, subjective, and prone to inter-observer variability. This paper presents a deep learning-based approach for automated pneumonia detection from chest X-ray images using transfer learning with pre-trained convolutional neural network (CNN) architectures. We evaluate the performance of three widely adopted models — ResNet50, VGG16, and DenseNet121 — on the publicly available Kaggle Chest X-Ray Images (Pneumonia) dataset containing 5,856 labeled images. The models are fine-tuned with data augmentation techniques to improve generalization. Our experimental results demonstrate that DenseNet121 achieves the highest classification accuracy of 93.27%, with a recall of 97.44% for pneumonia-positive cases, outperforming both ResNet50 (91.83%) and VGG16 (90.06%). The proposed framework offers a reliable, efficient, and scalable computer-aided diagnostic (CAD) tool that can assist radiologists in clinical decision-making, particularly in resource-constrained healthcare settings.

DOI: https://doi.org/10.5281/zenodo.19942378

 

Published by: vikaspatanker

Authors: Dr. Pravin Kumar, Abhinav Kumar

Abstract: This study evaluates the effectiveness of Janani Suraksha Yojana (JSY) in improving maternal and child health outcomes in Bihar. Using a mixed-method approach and regression analysis, the study finds that awareness and education significantly influence institutional delivery, while barriers such as cost and accessibility persist

DOI: https://doi.org/10.5281/zenodo.19693897

Published by: vikaspatanker

Authors: Aditiya Sawant, Devansh Sojitra, Rohan Maheta

Abstract: In modern educational environments, a significant number of students face learning difficulties due to slower cognitive processing, limited conceptual understanding, low confidence, and the inability of traditional teaching systems to address individual learning needs. These students, commonly referred to as slow learners, often require personalized academic support and innovative teaching strategies to achieve educational success. This paper presents Supporting Slow Learners with Remedial Innovation, an intelligent and inclusive educational support platform designed to enhance the academic performance and confidence of slow learners through customized remedial solutions. The proposed system integrates personalized learning plans, adaptive teaching methodologies, educator guidance, progress monitoring, and technology-assisted learning resources to create a supportive learning ecosystem. The platform allows educators to identify student weaknesses, assign customized study materials, track academic progress, and provide targeted interventions based on individual performance. Students can access simplified learning content, practice modules, motivational feedback, and continuous assessments according to their pace of learning. The system is developed using modern web technologies with a scalable frontend-backend architecture to ensure accessibility, usability, and performance. It supports multiple user roles including students, educators, and administrators for efficient management and monitoring. Experimental outcomes indicate improvements in student engagement, learning consistency, conceptual understanding, and confidence levels when compared with conventional classroom-only teaching approaches. This project demonstrates how remedial innovation combined with digital technologies can transform the educational journey of slow learners by promoting equal learning opportunities, reducing academic gaps, and creating an inclusive learning environment. The proposed framework can be extended in future with Artificial Intelligence, predictive analytics, and multilingual learning support for wider educational impact.

DOI: http://doi.org/

Published by: vikaspatanker

Authors: Jyoti Gupta, Ayush Patel, Siddharth Prabhudesai, Rahul Neve

Abstract: Customer churn prediction plays a vital role in helping businesses retain customers and minimize revenue loss in competitive markets. This study focuses on developing a predictive framework to identify customers who are likely to discontinue a service based on historical data. The dataset used in this project consists of customer demographic, behavioral, and financial attributes, which are preprocessed and transformed through feature engineering techniques to improve model performance. Multiple machine learning classification models are implemented and evaluated to determine their effectiveness in predicting churn. To address the issue of class imbalance, appropriate techniques are applied to ensure fair model training. The models are assessed using key performance metrics such as accuracy, precision, recall, F1-score, and ROC-AUC, providing a comprehensive comparison of their predictive capabilities. The analysis highlights the importance of factors such as customer tenure, service usage patterns, and billing characteristics in influencing churn behavior. The results demonstrate that machine learning models can effectively capture underlying patterns in customer data and provide reliable predictions. This study offers valuable insights into churn prediction and presents a data-driven approach that can support businesses in designing targeted customer retention strategies.

DOI: https://doi.org/10.5281/zenodo.19692500

Published by: vikaspatanker

Authors: Kanuri jai sai Prakash, Challa uday kiran, Gugilla Harshith, V.vidya sagar

Abstract: With the aid of a specially designed Graphical User Interface (GUI), a combination of Machine Learning and Deep Learning techniques was used to detect brain strokes. Images of "Stroke" and "Normal" cases were categorized from a dataset. Following the loading of the dataset, preprocessing and feature extraction were carried out, and then the data was divided into training and testing sets. The Convolutional Neural Network (CNN) algorithm achieved a significantly higher accuracy of 98% than the Support Vector Machine (SVM) algorithm, which only managed 59%. CNN outperformed SVM in stroke image classification, according to comparative analysis. The trained CNN model was then applied to new test image prediction, effectively differentiating between normal and brain cases. These findings demonstrate how well deep learning techniques work for precise Brain stroke detection from medical images. A crucial medical application that makes use of contemporary technologies like machine learning (ML) and deep learning (DL) for early stroke diagnosis and prediction is brain stroke detection. The automatic detection of ischemic and hemorrhagic strokes from CT and MRI scan images is the main focus of this study. Support Vector Machines (SVM), Random Forest (RF), and Logistic Regression are important algorithms for classification tasks. Images are classified, features are extracted, and stroke-affected brain regions are segmented using deep learning models, specifically Convolutional Neural Networks (CNNs) and Artificial Neural Networks (ANNs). Key procedures for the project include image enhancement, data preprocessing, and model training with frameworks like PyTorch, TensorFlow, or Keras. Metrics like accuracy, precision, recall, and F1-score are used to assess these models' performance. The accuracy of the model's stroke prediction is improved by adding clinical data, such as blood pressure, diabetes, smoking patterns, and other risk factors. Building an effective clinical decision support system that can aid in the early detection of strokes is the ultimate goal, as it may lower the death and disability rates related to cerebrovascular accidents (CVA).

DOI: https://doi.org/10.5281/zenodo.19691995

Published by: vikaspatanker

Authors: Noyal Biju, Dharunkumar C, Aziz Pardiwala, Abhishek Pillai

Abstract: Accurate and timely detection of brain tumors is a critical challenge in medical imaging, directly influencing treatment planning and patient prognosis. Conventional diagnostic approaches based on manual interpretation of Magnetic Resonance Imaging (MRI) scans are often limited by subjectivity, inter-observer variability, and increasing workload on radiologists. This study presents a robust deep learning–driven framework for automated brain tumor detection and classification, leveraging advanced Convolutional Neural Network (CNN) architectures. The proposed model employs a transfer learning approach using a pre-trained VGG16 network, fine-tuned on a curated dataset of MRI images to capture domain-specific features. A comprehensive preprocessing pipeline—including image normalization, resizing, denoising, and intensity standardization—is integrated with data augmentation techniques to address class imbalance and enhance generalization. The architecture incorporates fully connected layers with dropout regularization to mitigate overfitting and improve model stability Model performance is rigorously evaluated using standard metrics such as accuracy, precision, recall, F1-score, and Area Under the Receiver Operating Characteristic Curve (ROC-AUC). Experimental results demonstrate high classification performance, indicating the model’s capability to effectively distinguish between tumor and non-tumor cases. Furthermore, comparative analysis with baseline models highlights the superiority of the proposed approach in terms of feature extraction efficiency and predictive accuracy. The system offers significant potential for real-world clinical integration by reducing diagnostic latency, minimizing human error, and providing decision support for radiologists. This research underscores the transformative role of deep learning in medical image analysis and establishes a scalable foundation for future advancements, including multi-class tumor classification and explainable AI-driven diagnostics.

Published by: vikaspatanker