Real-Time Object Detection in Medical Imaging Using YOLO Models for Kidney Stone Detection (Published)
Kidney stone detection is essential for timely diagnosis and treatment, and the use of computer vision techniques has significantly improved this process. This study compares the performance of two advanced object detection models, YOLOv8 and YOLOv10, applied to kidney stone detection in CT scan images. YOLOv8, known for balancing speed and accuracy, incorporates the C2F building block for efficient feature extraction. YOLOv10 introduces NMS-free training, which eliminates the need for non-maximum suppression, resulting in faster inference and improved detection efficiency. We trained and evaluated both models using a dataset of annotated medical images, measuring their performance based on accuracy, precision, recall, and inference time. YOLOv10 outperformed YOLOv8 in terms of accuracy and precision, while YOLOv8 showed faster training convergence. The findings of this study provide valuable insights into selecting appropriate models for real-time medical imaging applications, depending on accuracy and resource requirements.
Keywords: : kidney stone detection, C2f, CT scan analysis, NMS-free training, medical image analysis, nephrolithiasis detection, real-time detection
Real Time Credit Card Fraud Detection and Reporting System Using Machine Learning (Published)
This study addresses the critical issue of real-time credit card fraud detection using machine learning. The primary goal is to develop a model that promptly identifies fraudulent transactions and alerts users. Two algorithms—Random Forest and Decision Tree Classifier were used, alongside various sampling techniques to balance the dataset and enhance performance. Six models were created, each with different accuracy levels in fraud detection. Key findings include a higher incidence of fraud among individuals over 75 years, likely due to less familiarity with modern transaction methods. Additionally, a majority of transactions involved females, indicating a potential higher fraud risk in these transactions. The Random Forest -SMOTE [Hyperparameter Tuned] model was the most effective, achieving a 97% accuracy rate, 95% F1 score, and 98% precision rate. For practical application, this model was integrated with Twilio for real-time fraud alerts, proving successful in sending timely, accurate notifications. The study highlights valuable insights and a robust solution for real-time fraud detection and response. Regular performance evaluations of the model are recommended to maintain its effectiveness against evolving fraud patterns.
Keywords: Algorithm, credit card fraud, machine learning, real-time detection, twilo integration