Category Archives: Uncategorized

Authors: Rajat Takkar, Disha Sharma, Hridyesh Sharma

Abstract: The rapid growth of edge computing has changed how artificial intelligence is deployed on devices with limited resources such as smartphones, embedded systems, and IoT devices. In such environments, constraints related to memory, power, and storage make it difficult to use traditional deep learning models directly. Although modern neural networks perform well in tasks like computer vision, they often need high computational resources, which limits their practical use on edge devices. In this work, we focus on lightweight deep learning architectures that are designed to operate efficiently under these constraints. Specifically, we examine three widely used models—MobileNetV2, SqueezeNet, and EfficientNet-B0—for real-time inference on edge devices. The CIFAR-10 dataset is used as a benchmark to evaluate model performance. To improve training efficiency, we also apply transfer learning by utilizing features from pre-trained models. In addition, optimization techniques such as structured pruning and dynamic quantization are used to reduce unnecessary parameters and improve computational efficiency without significantly affecting performance. These methods help in lowering model size and speeding up inference, making deployment more feasible in resource-limited environments. The experimental results show noticeable differences in performance across the selected models. EfficientNet-B0 achieves the highest classification accuracy of 92.06%, while SqueezeNet provides faster inference due to its compact architecture and fewer parameters. MobileNetV2 offers a balanced trade-off between accuracy and latency, making it suitable for practical applications. Overall, the findings highlight the importance of selecting appropriate lightweight architectures along with effective optimization strategies when deploying deep learning models on edge devices. This work provides useful insights into balancing accuracy, model size, and inference speed, which are key factors in real-world edge computing scenarios.

DOI: https://doi.org/10.5281/zenodo.19926432

Published by: vikaspatanker

Authors: Dr. Shrikant V. Sonekar, Professor Rohan B Kokate, Miss. Vaishnavi R Tandulkar

Abstract: The rapid growth in global energy demand, coupled with increasing environmental concerns, has accelerated the transition toward renewable energy sources, with solar power emerging as one of the most promising and sustainable alternatives. Despite its advantages, the efficiency and performance of solar power plants are significantly influenced by dynamic environmental conditions such as solar irradiance, temperature variations, dust accumulation, cloud cover, and equipment degradation over time. Traditional monitoring and control mechanisms are often reactive, manual, and incapable of handling large-scale data, resulting in suboptimal performance and increased operational costs. In this context, Machine Learning (ML) has gained considerable attention as a powerful tool for enhancing the efficiency and reliability of solar energy systems This paper presents a comprehensive study on the application of Machine Learning techniques to improve the efficiency performance of solar power plants. The proposed approach utilizes data-driven models to analyze historical and real-time data collected from solar panels, sensors, and weather forecasting systems. Various supervised learning algorithms, including Linear Regression, Random Forest, and Support Vector Machines (SVM), are employed for accurate prediction of solar power generation and identification of performance patterns. Furthermore, advanced deep learning models such as Artificial Neural Networks (ANN) are implemented to handle complex nonlinear relationships between environmental variables and energy output. In addition to energy prediction, the system incorporates intelligent fault detection and predictive maintenance mechanisms. Machine Learning algorithms continuously monitor system parameters to detect anomalies such as panel degradation, inverter malfunctions, shading effects, and wiring faults. Early detection of such issues enables timely maintenance, reducing downtime and improving overall system reliability. The integration of predictive analytics also allows operators to optimize panel orientation, tilt angles, and tracking mechanisms, thereby maximizing energy capture throughout the day. The proposed ML-based framework is evaluated using a dataset comprising solar irradiance, temperature, humidity, and historical power output records. Experimental results demonstrate a significant improvement in prediction accuracy and operational efficiency compared to conventional methods. The system achieves up to 20–30% enhancement in energy output efficiency, along with a considerable reduction in maintenance costs and system failures. Additionally, real-time monitoring and automated decision- making contribute to improved scalability and adaptability of solar power plants.

DOI: https://zenodo.org/records/19925745

Published by: vikaspatanker

Authors: Dr. Shrikant V. Sonekar, Professor Rohan B Kokate, Miss. Samiksha S Raut

Abstract: Cloud computing has revolutionized the way data is stored, processed, and shared by providing scalable, flexible, and on-demand access to computational resources over the internet. It has enabled individuals, enterprises, and government organizations to efficiently manage large volumes of data without investing heavily in physical infrastructure. Despite these advantages, the rapid adoption of cloud platforms has introduced significant challenges related to data security, privacy preservation, and fine-grained access control. Since data is stored on third-party servers, users lose direct control over their sensitive information, increasing the risk of unauthorized access, insider threats, and data breaches. Traditional encryption techniques such as symmetric and asymmetric cryptography ensure data confidentiality but fail to provide flexible and scalable access control mechanisms in dynamic, multi-user cloud environments. These methods rely heavily on complex key management systems and are not suitable for scenarios where access permissions need to be defined based on user roles, attributes, or contextual conditions. To address these limitations, Attribute-Based Encryption (ABE) has emerged as a powerful cryptographic approach that enables secure and flexible data sharing by enforcing access policies based on user attributes rather than identities. In particular, Ciphertext-Policy Attribute-Based Encryption (CP-ABE) allows data owners to define access structures directly within the encrypted data, ensuring that only users whose attributes satisfy the defined policies can decrypt and access the information. This paper presents the design and implementation of a secure and privacy-preserving data-sharing framework based on CP-ABE in cloud computing environments. The proposed system incorporates advanced security features such as fine-grained access control, secure key generation and distribution, user authentication, and protection against common attacks including collusion attacks and unauthorized data access. Additionally, privacy-preserving mechanisms are integrated to ensure that sensitive user attributes and data remain protected even from cloud service providers. The system architecture includes key components such as data owners, attribute authorities, cloud servers, and data users, working together to provide a secure and efficient data-sharing environment. Experimental evaluation demonstrates that the proposed framework significantly improves data security, reduces the risk of data breaches, and enhances access control efficiency compared to traditional encryption-based systems.

DOI: https://zenodo.org/records/19924638

Published by: vikaspatanker

Authors: Sourabh Vijay Patil, Vaishnav Maruti Kadam, Ajay Angad Ahir, Altaf Yasin Mahat

Abstract: Missing person cases are a global concern that cause emotional distress for families and challenges for law enforcement agencies. Traditional search methods such as posters, manual surveillance, and public announcements are slow and inefficient. This paper proposes FaceTrace, an artificial intelligence based missing person detection system that uses deep learning facial recognition to identify individuals from images and surveillance streams. The system leverages ArcFace embeddings, computer vision techniques, and a centralized MySQL database to match uploaded images with stored records. The proposed system enables faster identification and improves accuracy compared to manual methods.

DOI: http://doi.org/

Published by: vikaspatanker

Authors: Manish S. Bansode, Tejas S. Mokal, Saad S. Pathan, Karan K. Rathod, Professor Yash S. Shet, Professor Hemanth K.Thakur, D. N. Jaiswal

Abstract: The rapid increase in construction activities has significantly increased the demand for water used in concrete curing. Conventional curing methods require continuous external water supply, which is often impractical in regions with water scarcity. This research focuses on self-curing concrete using Polyethylene Glycol (PEG-400) and Polyvinyl Alcohol (PVA) as internal curing agents. The study evaluates the mechanical properties of concrete, particularly compressive strength, by varying the percentage of these agents. The results demonstrate that self-curing concrete improves hydration, reduces shrinkage, enhances durability, and minimizes water consumption. The study concludes that PVA shows better performance compared to PEG in terms of strength and water retention.

DOI: http://doi.org/

Published by: vikaspatanker

Authors: Udit Tripathi

Abstract: SocioMind AI is an AI-powered analytical framework that quantifies psychological states and personality traits through the automated processing of heterogeneous social media data. Unlike traditional sentiment analysis — which reduces complex human communication to a single positive/negative polarity score — SocioMind AI employs a multi-dimensional approach to construct a comprehensive "Linguistic DNA" profile of an individual, correlating public persona signals with private aspirational data to deliver a 360-degree behavioral footprint. The system operationalizes a novel concept: the Digital Behavioral Footprint (DBF) — the aggregate, cross-contextual trace that an individual leaves across multiple social media channels, each reflecting a different facet of their psychological identity. By processing and cross-referencing Primary Content, Interactional Tone, Interest Graphs, and Aspirational Signals simultaneously, SocioMind AI achieves what single-channel sentiment tools cannot: a holistic, internally-validated psychological portrait. At its inference core, SocioMind AI leverages the Gemini 3 Flash large language model architecture, optimized for structured JSON output to ensure deterministic, research-grade data handling. The analytical output spans Big Five personality trait quantification, Emotional Density Mapping, and derived psychological indicators including Social Stress Levels, Behavioral Consistency scores, and Mood Trajectory projections. The system is implemented as a React-based web application with Recharts-powered radar and bar chart visualizations, making complex psychological matrices accessible to both researchers and non-specialist users. Validation experiments across 300 profiles demonstrate Cohen's kappa = 0.74 for Big Five dimensions and Pearson r = 0.81 for emotional valence detection, establishing SocioMind AI as a viable zero-knowledge psychological proxy for research-grade personality inference.

DOI: http://doi.org/

Published by: vikaspatanker

Authors: Daksh Jasrotia, Associate Professor Dr.Simarjit Kaur

Abstract: Public Wi-Fi networks in places like airports or cafés is convenient, but they are not very safe. Most people do not realize that these networks don’t usually have strong security and because of that, attackers can use methods like Man-in-the-Middle (MITM) attacks. One common method of MITM attacks is called ARP spoofing. In this method, a hacker tricks your device into sending data through their system instead of the real network. This usually happens in the background and the users don’t notice anything suspicous. As a result, important and personal information can be exposed and stolen without the user knowing. In this paper, I am proposind WiFiShield, a lightweight system or rather an application designed to detect such vulnerabilities in real time. Instead of being a heavy security suite, the application or system runs quietly in the background, sniffing network packets and double-checking the ARP table for any sudden, suspicious changes or spikes in the ARP requests. If the system sees a gateway address "flapping" or changing unexpectedly, it assigns a risk score and alerts the user immediately. I am aiming to make WiFiShield highly effective at spotting these hijacks without slowing down the computer. The application should be a simple, low-power solution for anyone who needs to stay connected on the go without leaving their personal data wide open to hackers.

DOI: http://doi.org/

Published by: vikaspatanker

Authors: Akash Suri, Aryan Pathania, Divyayush Verma, Rajat Takkar

Abstract: The proliferation of fake news on social media plat- forms poses significant threats to public discourse and democratic processes. While numerous machine learning approaches have been proposed for fake news detection, limited attention has been given to understanding why different models classify news as fake and whether these explanations are consistent across algorithms. This paper presents a comparative and explainable machine learning framework that addresses two critical research questions: (1) Do different ML models agree on which textual fea- tures indicate fake news? (Trust Gap Analysis), and (2) Do fake news patterns learned from one domain generalize to another? (Cross-Dataset Robustness). We evaluate four classical machine learning algorithms—Logistic Regression, Naive Bayes, Support Vector Machine, and Random Forest—using TF-IDF features on two distinct datasets: ISOT (political news, 44,898 articles) and WELFake (general news, 72,134 articles). Using SHAP (SHapley Additive exPlanations) for model interpretability, we compute Jaccard similarity and Spearman rank correlation to quantify agreement between model explanations. Our results reveal that different models exhibit varying levels of agreement on fake news indicators, with implications for model selection in real- world deployment. Furthermore, cross-dataset analysis identifies “universal” fake news features that generalize across domains versus “topic-specific” features that are domain-dependent. This work contributes a novel analytical framework for evaluating the trustworthiness and generalizability of fake news detection systems.

DOI: https://doi.org/10.5281/zenodo.19916551

Published by: vikaspatanker

Authors: Professor Rajendra Pawar, Omkar Walunj, Pranav Hole, Sarthak Thigale, Sohan Sandbhor

Abstract: Early detection of brain tumors is crucial for effective treatment and improved patient outcomes. This study presents an automated system for brain tumor classification using deep learning techniques. A convolutional neural network based on the VGG16 architecture is utilized to analyze MRI images and classify them into different categories such as glioma, meningioma, pituitary tumor, and normal cases. The system includes image preprocessing, model prediction, and a web-based interface developed using Flask for easy user interaction. Users can upload MRI images and receive instant predictions along with confidence scores. Additionally, a PDF report is generated to present the results in a structured format. The proposed approach demonstrates reliable performance and can assist medical professionals in making faster and more accurate preliminary diagnoses.

Published by: vikaspatanker

Authors: Kavya Sharma

Abstract: The residential property market in Noida is complex due to its structured sector-based planning and the coexistence of Authority-developed plots and private high-rise housing societies. These two categories follow different pricing patterns, even within nearby areas. This study aims to develop a transparent price prediction model using Multiple Linear Regression to analyze the impact of hyperlocal features, particularly Metro connectivity, on property prices. A historical dataset of Noida properties was utilized and processed using Python and Pandas. The finalized regression model achieved approximately 85% accuracy on the testing dataset, revealing that Sector Location and Metro Connectivity are the most influential factors, often outweighing flat size. This demonstrates that a transparent regression approach can effectively support fair pricing in high-variance markets.

DOI: http://doi.org/

Published by: vikaspatanker