Associate Professor and Head, Department of Artificial Intelligence and Data Science, Vidyavardhini's College of Engineering and Technology, Maharashtra, India
Thursday August 27, 2026 3:28pm - 3:30pm IST Virtual Room FGOA, India
Authors - Piyali Karmakar, Pabitra Mitra, Manjira Sinha Abstract - Communication is fundamental to human connection. Individuals with complex communication needs (CCN), such as those with cerebral palsy, often face significant barriers to speech and language expression. Augmentative and Alternative Communication (AAC) systems address these challenges through the use of graphical symbols. In this work, we strengthen AAC capabilities by creating specialized datasets that support bidirectional translation between symbolic language and natural English text using NLP techniques (Sym2NL). The datasets are enriched with tense and narrative features to improve contextual accuracy.We also present PictoGen, a text-to-picture generation module designed to visually represent unfamiliar words or concepts. Together, these contributions support more natural, expressive, and accessible communication.
Authors - Pallavi Patil, Mansing Rathod Abstract - The rapid evolution of malware, including polymorphic and fileless variants, has weakened traditional detection methods. This paper looks at sophisticated malware detection frameworks that use deep learning and machine learning to assess important malware features such network anomalies, opcode sequences, and API requests. Signature-based techniques are effective at identifying known dangers, but they are not very effective at thwarting zero-day assaults. While they offer improvements, alternative strategies including behavior-based, cloud-based, and deep learning techniques also have drawbacks. The current detection frameworks unify real-time threat information with two IDS detection approaches to enhance security capabilities. The review extends its analysis to model interpretability and evaluates the computational burden. The assessment of experimental findings helps researchers enhance adaptable malware security through the display of improved detection precision and resilient capabilities versus evolving cyber threats
Authors - Pragathi Guduru, Ramya S, Anitha H Abstract - Hand sign recognition systems play a crucial role in bridging communication gaps for people with hearing and speech impairments. This review paper explores various methodologies and algorithms employed in previous research on hand sign recognition, analyzing their performance, accuracy, computational efficiency, and effectiveness in real-world applications. Special emphasis is given to algorithms related to the Discrete Fourier Transform (DFT), including the Hebbian Classifier, Radial Basis Function (RBF) networks, and Self-Organizing Maps (SOMs), which have been utilized for feature extraction, pattern recognition, and classification. The study also examines deep learning approaches such as Convolutional Neural Networks comparing their strengths and limitations. Additionally, the paper highlights how these advances contribute to assistive technologies in healthcare, aiding doctors during medical procedures, and improving accessibility for individuals in need. By providing a comparative analysis of these techniques, this review aims to offer insights into the most effective strategies for enhancing hand sign recognition systems, paving the way for future research and innovation in the field.. . .
Authors - Ajay Talele, Sujal Tawale, Tushar Ghorpade, Nikhil Wagh, Aryan Sable, Pallav Vaniya, Yashashree Mehare, Parishnav Thokal, Samruddhi wayal, Vedant Motale Abstract - Book discovery in the digital age is extremely difficult which is the result of various factors such as users struggling with information overload as well as trying to find the content most closely to their needs. Albeit there are many recommendation systems available in the market, most of them are just based on general ratings and neither do they use the rich metadata that is available from external book sources nor do they provide the functionality of exploring related work in the best way. Through the use of our proposed book recommendation system, the constraints that are in place currently can be very easily overcome effectively by the use of more advanced reinforcement machine learning and data integration techniques. What the model would do is to analyze users' reading history and preferences and combining data from bookstores so that an efficient and effective model would be built which would give accurate suggestions of books to the users according to their preference. The Python language is chosen as the basis for development and for the backend, the Flask framework is used while for finding the most appropriate document for the reader, TF-IDF vectorization, and cosine similarity are employed. Moreover, the linkage of outside APIs not only makes it more in-depth to look at but also increases the system's accuracy. Our approach enables the users to discover new and personalized books in a simple and efficient manner. Our project is a direct contribution to a highly interactive reading journey and it also contributes to increasing love for literature by giving the users the opportunity to find books that will truly engage them.
Authors - Mariya Joseph, Vinod Kumar K Abstract - Greenwashing, the practice of brands making false or exaggerated environmental claims, has become a major concern across industries, especially in the food, fashion, and beauty sectors, as consumer demand for sustainable and ethically made goods rises. This study examines how consumers react to greenwashing, with a particular emphasis on their awareness, attitudes, and behaviors in the face of false sustainability promises. This study investigates how consumers recognize and interpret greenwashing, the emotional and cognitive elements affecting their reactions, and the actions they take in response—such as boycotting brands or looking for more transparent alternatives—by analyzing consumer surveys and existing literature. The paper also delves into the role of brand trust, social media, and regulations in shaping consumer reactions to green-washing. Results indicate that although consumers are become more conscious of greenwashing, there is still a sizable gap in their capacity to recognize false claims. The study emphasizes how crucial third-party certification, brand openness, and consumer education are to reducing the damaging effects of greenwashing. In the end, the study urges consumers and brands to take a more proactive and knowledgeable stance inorder to guarantee that sustainability initiatives are sincere and significant.
Authors - Ajay Talele, Omkar Shinde, Shrey Rai, Shreyash Mutha, Rohan Shelke, Sumedh Malode, Soyam Maykar, Siddhesh Manjare Abstract - With the increased demand for secure and efficient ways to transfer files, peer-to-peer systems have developed as good solutions. Seen with client-server systems, they can be very successful, however, typically suffer from bottlenecks and single points of failure, along with being less efficient for large quantities of data exchange - P2P networks tend to instead successfully allocate workloads over different peer nodes and distributed information along with concurrent and fault-tolerant characteristics. This paper details the design and implementation of a multi-threaded file-sharing system using Java Sockets, multi-threading concepts, and other relevant networking ideas. This system allowed multiple users to exchange files in a secure way over a network efficiently, allowing for concurrency in this system through efficient thread synchronization. Each peer operated independently as both a client and a server, allowing for collaboration and facilitated file transfers across nodes without a central authority.
Authors - Ayushi Sensharma, Ashish Sharma, Swati Agrawal Abstract - The gender discrimination is a significant issue in the labor market. “Motherhood Penalty” is one of the important contributors to this issue. This study aims to find the evidence of impact of parenthood on employment to population ratio and mean nominal monthly earnings concerning factors – household structure and number of children under age six. Using interactive multiple linear regression models, we have derived meaningful conclusions from data collected from the International Labor Organization (ILO). Our findings reveal that there is a significant motherhood penalty in India. Women’s employment probability decreases by 12.4% with one child and up to 19.09% with three or more children. Meanwhile, men experience a fatherhood bonus, with employment rates rising by up to 24.79% as they have more children. Wage disparities are also evident—mothers with two or more children earn substantially less than childless women, whereas the fatherhood wage premium is weaker than in developed economies. Wage disparities are also evident. Mothers with two or more children earn substantially less than childless women, whereas the fatherhood wage premium is weaker than in developed economies. Through this study, we also see the probable reasons behind the results observed from the models. Lack of institutional support for working moms, workplace prejudice, and deeply rooted gender stereotypes are some of the main reasons attributing to the “Motherhood Penalty”. This disparity is further exacerbated by strict work rules, poor childcare facilities, and lax paternity leave regulations. Overall, the motherhood penalty is a serious phenomenon affecting the lives of many mothers and degrading their standards of living.
Thursday August 27, 2026 3:30pm - 5:30pm IST Virtual Room FGOA, India
Authors - Rupali Parte, Vaishali Kapure, Pranav Bankar, Shrutika Mandharne, Avadhoot Khandagale Abstract - This research presents a cloud-integrated machine learning (ML) system designed to enhance e-commerce operational efficiency. Leveraging Microsoft Azure’s data services (Azure Data Factory, Databricks, ADLS Gen1 and Gen2, and Power BI), along with a Streamlit user interface, the system processes large-scale transactional data for real-time analytics and decision-making. Four specialized ML models address key challenges: logistics clustering optimizes shipping routes; sales forecasting improves inventory management; fraud detection strengthens security; and order cancellation prediction enhances customer retention. The automated data pipeline ensures efficient ingestion, transformation, and storage, minimizing latency and maximizing data accessibility. The interactive Streamlit interface allows users to select and deploy models, while Power BI dashboards provide dynamic visualizations. This integrated approach demonstrates the potential of cloud computing and ML to improve logistics, enhance fraud prevention, and optimize revenue forecasting.While offering scalability, the system necessitates robust security measures to address data privacy concerns. The reliance on historical data also necessitates continuous model monitoring and retraining to mitigate potential biases. This research contributes a practical framework for e-commerce businesses seeking to leverage data-driven insights for improved performance.
Authors - Vangala Thanusree, Rekapalli Bhagya Srilakshmi, Kanikireddy Harshitha, Sushama Rani Dutta, A. Pranathi, Boga Sudharshini Sree Abstract - Crime is a persistent social issue that impacts public safety and urban development. With the rise in data availability and machine learning techniques, predictive modeling of crime rates has become a valuable tool for law enforcement and policy planning. We suggest a combination machine learning strategy in this paper that integrates both spatial and temporal data, alongside social and economic indicators such as decographic and Economic Indicators rate, literacy, and income levels, to enhance crime rate prediction accuracy. We evaluate the effectivness of multiple models, including RF, XGBoost, and a hybrid ensemble of both, on a real-world dataset comprising crime statistics from multiple Indian states. Our results demonstrate that integrating socio-economic factors significantly improves model performance, offering deeper insight into crime patterns and enabling data-driven intervention strategies. The proposed model outperforms traditional single-model baselines, achieving higher accuracy and F1 scores across various crime categories. This approach serves as a robust framework for smart policing and proactive crime prevention in high-risk zones.
Authors - Varun Mittal, Madhan Kumar Srinivasan Abstract - Artificial Intelligence (AI) is revolutionizing industries with its capabilities in automating tasks, enhancing decision-making, and providing predictive insights. A clear way to frame the current state of AI is to acknowledge that it’s still a technology. To fully leverage its benefits, whether for business or personal purposes, one must understand and learn to use it effectively and adapt workflows to align with its strengths and weaknesses. Organizations all around the world are transforming their existing systems and building new systems to leverage the power of artificial intelligence, but these advancements to enhance their businesses come with significant security challenges. These security threats pose a challenge to both the service providers (developers) as well as the customers. This paper delves into the security issues within AI that organizations and their users can face with AI systems, categorized under state-of-the-art AI security taxonomies.
Authors - Santhameena S, Shaunak Agrawal, Shobith R Prabhu, Shaurishail M Awanti, Siddharaj Dhegaskar Abstract - This work focuses on military vehicle detection using Synthetic Aperture Radar (SAR) images from the MSTAR dataset. Challenges such as speckle noise, limited data size, and classification accuracy are addressed using preprocessing techniques, dataset augmentation via Spectral Normalization GANs (SN-GANs), and a custom-designed Convolutional Neural Network (CNN). The proposed methodology achieves an accuracy of 98.1%, showcasing the potential of GAN-augmented SAR datasets in target recognition tasks.
Associate Professor and Head, Department of Artificial Intelligence and Data Science, Vidyavardhini's College of Engineering and Technology, Maharashtra, India
Thursday August 27, 2026 5:30pm - 5:32pm IST Virtual Room FGOA, India
