Optimizing urine protein detection accuracy using the K-nearest neighbors algorithm and advanced image segmentation techniques

Abstract Kidney metabolic disorders are diagnosed by assessing protein levels in urine, which reflect renal health. Traditional detection methods are time-consuming and expensive. This study explores using the K-nearest neighbors (KNN) algorithm combined with advanced image segmentation for accurate urine protein detection. The research utilized a dataset of protein-level images captured by an ELP-type digital camera sensor, classifying them based on red, green, and blue (RGB) values. The KNN algorithm was tested with various K values (K = 3, K = 10, K = 20). Results showed that K = 3 provided the highest accuracy at 96.7%, with precision, recall, and F1-score of 97.0%, 96.7%, and 96.2%, respectively. Higher K values decreased accuracy, with K = 10 at 86.7% and K = 20 at 76.7%. These findings demonstrate that KNN can effectively predict protein levels, offering a promising and efficient alternative to traditional methods. The study also presents a prototype design for this detection approach.