Authors - SnehalBalasaheb Salve, Harsha Bhute Abstract - Developing a precise and strong model for forecasting energy utilization is importantaim for the management and functionality of smart buildings. The previous studies have researched different models for forecasting various load prediction schemes. The combined effects regarding data enrichment and machine learning approach in energy predictions have not been fully examined. This research proposes a novel approach, an ensemble model enhanced by generative adversarial networks (GANs) for predicting the usage of energy in big buildings that are commercial. This combined system integrates various single models using ensemble method with stacking. Furthermore, a GAN is utilized to capture the distribution of samples from the main dataset, generating top-notch specimens to augment the dataset from the training data. This expanded dataset allows the model to train with a wider range of samples, increasing its resilience. The experimental series evaluate the method that is proposed, using three variants of GAN and assessing performance with metrics such as mean absolute error, root mean square error, and coefficient of variation of root mean square error. This proposed approach demonstrates practical results that develop a model for power utilization prediction in application of real world.
Authors - G Roopa, H.L.Suresh Abstract - The work outlined here provides a new method to handle fault of Power Electronic Traction Transformer (PETT) switch using reverse charging Cascaded H Bridge (CHB) and Dual Active Bridge (DAB) topologies. Precisely, the main purpose is to improve the speed of detection, determining the location, and recovery of faults from the existing system using feature extraction and Machine Learning algorithms. The traditional approaches in achieving fault tolerance are defective in detecting faults in good time, isolating faults inadequately, and using backup hardware. In order to solve these problems, the proposed methodology actively reassigns control signals to backup modules resulting in the exclusion of faulty elements while preserving a stable system performance with moderate loss in efficiency. The feasibility of the suggested approach is confirmed through simulation outcomes for fault detection precision, which is increased to 98 percent; the fault localization time of at most 5-10 ms; and system throughput of 5-8 percent. Furthermore, the work investigates how CHB and DAB function in fault conditions and enshrine a novel reverse charging method for maintaining the DC voltage of the redundant module. The startup process of the PETT system is also managed with optimization of voltage and transient, which leads to enhance the general system initialization. Besides increasing the dependability and fault tolerance of PETT systems, the above methodology also reduces the system’s embedded hardware duplication and elevates system performance and scalability , which consequently leads to the decrease of the total system cost by 15 percent. These results point out that the proposed solution has potential for the development of the next generation of fault-tolerant power electronic systems.
Authors - Santushti Betgeri, Rohit Rathod, Sakshi Rathod, Sanskar Raut, Anisha Sadanshiv Abstract - AUTOHUB is an integrated solution for essential services regarding vehicles, as well as vehicle expenses. This is a solution with both a native Android app and the response web interface that can easily integrate, especially with dynamic time slot selection for service bookings and an extensive expense tracker for fuel, repairs, tolls/fines, and others. It also has a cloud Online Document Manager for safe document storage, automatic expiration reminders, different dashboards for service providers, and more. It is developed using Android Studio, React, Node.js with Express, Firebase Firestore for real-time data sync, and Razorpay for secure payment processing. The platform has a modular microservices architecture and is scalable and easy to maintain. This integrated solution not just tackles the current challenges posed by fragmented automotive service management, but it also builds the base upon which future extensions can be realized, placing AUTOHUB well ahead of its time as a platform for automotive care.
Authors - Ashwini Matange, Harsha Talele, Pratik Nagare, Vineet Morankar, Aniket Gavkare, Moin Shaikh Abstract - Accurate segmentation and prognostication of brain tumors are critical for effective diagnosis, treatment planning, and patient management in glioma. In this work, we present a unified framework built upon the BRATS2020 challenge data that integrates deep learning-based segmentation with radiomics and machine learning for overall survival prediction. First, we employ a 3D-UNet architecture to perform robust segmentation of brain tumors from multi-modal MRI scans, achieving a mean Intersection over Union (IOU) of 86%. This segmentation not only delineates tumor sub-regions effectively but also provides the basis for subsequent feature extraction. Leveraging the pre-trained 3D-UNet, we extract deep features from the MRI scans, and in parallel, perform radiomics feature extraction on the corresponding tumor masks. These features are then combined with clinical and demographic data provided in the BRATS2020 challenge dataset. A random forest classifier is subsequently trained on this comprehensive feature set to predict overall patient survival, achieving a classification accuracy of 70% in stratifying patients into survival categories. Our approach builds on recent advances in brain tumor segmentation—incorporating ideas such as ensemble learning, multi-modal imaging, and uncertainty quantification—to enhance both the segmentation accuracy and prognostication performance. The promising results demonstrate that the integration of deep learning segmentation with radiomics and traditional machine learning methods can serve as a robust tool for personalized treatment planning and risk stratification in glioma patients.
Authors - Vaishnavi Moorthy, Jagadeesan Moorthy, Shubhradip Saha, Anshuman Kumar Abstract - In the medical field, the readability of important information on medicine packs, like the expiry date, is of prime importance for maintaining patient safety. Yet, a number of reasons like damage, blurring, and printing defects may hide this important information on medicine strips. To solve this problem, we suggest a deep learning-based solution for medicine strip denoising and enhancement, making important information such as expiry dates more legible. Our approach utilizes image denoising, specifically designed to correct blurry or partially readable expiry dates on the packaging of medicines. This solution not only helps healthcare workers and patients validate medicines but also makes a contribution to the pharmaceutical industry.
Authors - Shweta Kumar, Saru Dhir, Ashish Kumar Mourya Abstract - Healthcare informatics has many difficulties due to the complexity of varied medical data. Clinical notes, imaging, and genomic data are instances of unstructured data that is more flexible and has more depth than organized data, such as digital records, which are easier to use. Combining different healthcare data sources is difficult due to interoperability issues and semantic variability. Despite the emergence of standardization projects such as HL7 (Health Level 7) FHIR (Fast Healthcare Interoperability Resources) and SNOMED CT (Systematized Nomenclature of Medicine Clinical Terms), inefficient processes and unreliable vocabulary continue to impede seamless communication of information. The dispensation of natural language, or NLP (Natural language processing), methods enable the extraction of important information from uncontrolled health information. Furthermore, instantaneous data analysis and scalability are enhanced by online computing, and blockchain technology is being investigated as a safe, independent method of sharing medical data. This study examines the challenges of managing a variety of healthcare information as well as the possible benefits of contemporary technologies. Future research focuses on improving interoperability frameworks, developing AI-driven data analysis, and ensuring confidentiality and security of data in order to provide effective and data-driven healthcare options.
Authors - Arun N, Rithika J Prabhu, DHANYA M Abstract - Sustainability in India has been become a driving force behind the growth of the green energy sector and the economy's transition to cleaner energy. The research paper investigates the use of machine learning models to predict stock prices of green energy companies in India. It deliberates on the rapid growth of the green energy market and the potential for ever-advancing technologies making accurate prediction in finance for supporting the nation's sustainable development goals. Using machine learning, it generates useful insight for stock performance for the benefit of investors and policy makers in arriving at decisions.
Authors - M. Chaitanya Raju, Maddu Reshma, V. Anvesh, Lekha S. Nair Abstract - Waste classification and management are important for healthier planet Earth. In this paper we are proposing an integrated approach for waste detection and classification using object detection along with natural language processing (NLP) techniques. which introduce a YOLO-based model to detect and classify waste in images by using Bootstrap Language-Image Pretraining (BLIP) for scene understanding and contextual analysis. The workflow involves, feeding the waste images into a preprocessing stage (image), captioning image data with Natural Language Processing (NLP) to produce descriptive captions, and analyzing the textual features of detected captions that exist in the waste (waste elements). The classification of the detected object is performed by a custom trained YOLOv8 model which is fine-tuned on a specific waste class dataset. Experiments show that the model recognizes garbage, recyclables and litter with high accuracy. This system showcases the potential of combining visual and textual modalities to enhance waste detection accuracy, offering a robust tool for automated environmental monitoring and management.
Authors - Juttiga Rohita, B Teja Sree, Ibrapatnam Anusha, Mohammad Sharmila Begum, Nirjogi Mahathi Abstract - Social media platforms have become increasingly vulnerable to online threats, making safeguarding the internet an increasingly difficult task. Why? This project showcases an artificial intelligence-powered system that can detect and filter out inappropriate text and images in real-time. Machine learning and natural language processing (NLP) are utilized by the system to detect hate speech, toxic terminology such as slang, and explicit imagery while maintaining document integrity. TF-IDF, LSA, and Word Embeddings are utilized in text filtering to improve the understanding of context. In image filtering, deep learning models using convolutional neural networks (CNNs) and pre-trained NSFW classifiers detect and remove explicit content. This balances scale with accuracy and provides a robust, automated content moderation system that improves both safety and compliance on the Internet.
Authors - Shahedhadeennisa Shaik, Abhinav R B, Chaitra S P, Sagari S M Abstract - Video traffic surveillance has become an essential tool for various applications, including security, transportation planning, and traffic management. Recent advancements in deep learning have opened new possibilities for enhancing the performance of vehicle detection and tracking in these systems. This paper addresses the challenges of online action detection in surveillance scenarios by focusing on enhancing multi-object tracking (MOT) performance. Recognizing the limitations of current MOT methods in handling real-world surveillance complexities, we propose a methodology that integrates appearance model extraction directly from the object detector, adaptive adjustments of confidence thresholds and input resolutions, and the incorporation of color information into ReID embeddings. We aim to bridge the gap between motion-based and ReID-based tracking methods, improving both speed and accuracy. Our proposed techniques, including scene-based and object-based adaptation through reinforcement learning, and advanced feature fusion for ReID, are designed to enhance robustness and efficiency. We evaluate our methodology using publicly available datasets, focusing on surveillance-specific challenges. The enhancement in MOT performance is challenging and paving the way for more reliable and efficient surveillance system.
