Authors - Rupan Panja, Rajani K. Mudi, Nikhil R. Pal Abstract - The Support Vector Machine (SVM) is a popular classification algorithm, however, it suffers from the drawback that the classification time for an unknown data point is proportional to the number of support vectors (SVs). Thus, its application for real-time decision-making, especially for complex class boundaries becomes problematic / impossible. In this article, we propose an algorithm, Self Organizing Support Vector Machine (SO-SVM), which can decrease the number of SVs without compromising the accuracy. In this algorithm, we first cluster the data points using self-organizing map to find potential class boundary points, which are crucial for determining the separating hyperplane in an SVM. The separating hyperplane is then learned from the selected boundary points, leading to a reduction in the number of SVs. Learning the SVM also becomes efficient because of the reduced number of data points. The proposed algorithm has been tested on a number of benchmark datasets and found to decrease the number of SVs without deteriorating the classification accuracy. The SO-SVM algorithm thus can be an efficient alternative to the SVM algorithm and thus can be applied instead of normal SVM without making a noticeable compromise with the performance.
Authors - Nitesh Pradhan, Aryan Baghla Abstract - Electrocardiogram (ECG) signals are effective indicators for detecting obstructive sleep apnea (OSA) due to their ability to reflect physiological changes associated with apnea events. EfficientApneaNet is a deep learning-based model for the detection of OSA from a single-lead ECG. In general, traditional approaches are reliable but exhaustive and costly; therefore, ECG-based methods are being sought. Earlier machine-learning methods, including Support Vector Machines and Random Forests, are prone to real data noise. Based on these, EfficientApneaNet joins previous advances in deep learning for further improvement in accuracy and robustness. The proposed architecture is powered by three novelties, namely: the ENBlocks, inspired by EfficientNet, Squeeze-and-Excitation blocks, and attention. ENBlocks make use of depthwise separable convolutions that reduce the computational complexity and amplify the efficiency of spatial feature extraction. The Squeeze and Excitation blocks carry out channel recalibration to focus on relevant patterns, while the attention mechanism underlines the critical temporal events within the ECG sequences. On the Apnea-ECG dataset, EfficientApneaNet realizes state-of-the-art performance with 91.47% accuracy, 85.92% sensitivity, and 94.91% specificity outperforming those of leading existing CNN-LSTM hybrids. It adopts Adamax optimization for stability while implementing the technique of cosine annealing for LR scheduling. The residual connections avoid gradient vanishing and explosion. Through ablation studies, it is confirmed that SE blocks and the attention mechanism are both essential to achieving high sensitivity and high specificity. In this respect, EfficientApneaNet would be considered a significant improvement in OSA detection, as it has successfully handled spatial and temporal complexities in ECG data.
Authors - Arun Kumar S, Niveditha S, Vikas K B, Varshini C, Hemalatha H U Abstract - Improved patient outcomes and effective management of Chronic Kidney Disease (CKD) depend on early detection, making it a major worldwide health concern. This study presents a hybrid machine learning model that combines Random Forest and LightGBM classifiers in order to accurately predict CKD stages. The dataset from Kaggle was used to build the model, and SMOTE was used to handle class imbalances in addition to feature engineering and data preprocessing. With regard to test data, the model's accuracy was 97.7%. Enhancing its actual clinical utility, a web-based tool was also developed to enable real-time forecasts of CKD stage based on patient inputs. Using a mixed-method approach that included more than 1600 participants' medical examinations and surveys With the integration of real-time clinical data and strong predictive models among its potential enhancements, the proposed approach provides a robust and easily utilized tool for clinical applications.
Authors - Nilesh Korade, Swaraj Waykar, Archit Waghmode, Shweta Tate Abstract - Though solo travel can be an enriching and adventurous experience, staying alone (without social interactions) can sometimes enhance feelings of loneliness and this can be detrimental to a traveler's mental health. In this paper, we present an AI Travel Buddy. This intelligent system uses real-time facial expression recognition to determine the emotion of travelers at the moment and provides personalized conversational topics based on the detected emotion using the emotion-aware voice assistant. Using the latest CNN architecture for emotion detection, and Conversational model for dynamic answers, the AI Travel Buddy is a companion to travelers, delivering emotional support to users on solo trips. The paper describes the technologies used, the system architecture, and its methodology, showing substantial improvement in user engagement emotionally alongside providing overall higher travel satisfaction.
Authors - Sunil Sangve, Rutuparn Kakade, Saamya Gupta, Sahil Akalwadi, Soham Pawar Abstract - The paper introduces a novel and detailed approach for automated generation of fully customizable and user-centric food recipes. The system utilizes fine-tuned GPT2 and fine-tuned YOLOv9 to work efficiently. The user can input the available ingredients by uploading an image which is then processed by the fine-tuned YOLOv9 model to extract the ingredients present in the image. Alternatively, the user can also input ingredients manually. The user input also contains macro and micro nutrient levels, caloric values, cholesterol levels, taste and texture. Utilizing fine-tuned GPT2 LLM, the system then generates a unique recipe that follows the input. Web application developed by Next JS 14 with backend support from Flask, helps greatly in enhancing User experience. The system also utilizes Stable Diffusion to generate colorful recipe images for user reference of how the final recipe should look like.
Authors - Austine S Manuel, Alana Ance John, Hamna Rafeeq, Thalhah Anas, Manoj V. Thomas Abstract - Smart assistive technologies improve the independence and access to transportation of the vision-impaired people [1]. This paper presents AI-enhanced smart glasses designed to empower visually impaired by transforming their interaction with the environment. The system combines cutting-edge technologies to offer a comprehensive assistive solution. A path navigation algorithm, powered by YOLOv8, ensures safe mobility by detecting obstacles and guiding users with audio feedback. Integrated OCR capabilities enable real-time text reading, while an image captioning module provides detailed scene descriptions for enhanced situational awareness. The glasses incorporate a camera and an ultrasonic sensor, delivering robust performance in diverse scenarios, including detecting objects at close proximity where traditional algorithms fall short. With seamless audio output for user interaction, the proposed solution bridges the gap between technology and accessibility, promising independence and confidence for visually impaired individuals. This innovative fusion of AI and sensory inputs defines a new benchmark for assistive devices.
