Utilizing machine learning algorithms for precise discrimination of glycosuria in fluorescence spectroscopic data

Abstract

Fluorescence spectroscopy coupled with a random forest machine learning algorithm offers a promising non-invasive approach for diagnosing glycosuria, a condition characterized by excess sugar in the urine of diabetic patients. This study investigated the ability of this method to differentiate between diabetic and healthy control urine samples. Fluorescent spectra were captured from urine samples using a Xenon arc lamp emitting light within the 200 to 950 nm wavelength range, with consistent fluorescence emission observed at 450 nm under an excitation wavelength of 370 nm. Healthy control samples were also analyzed within the same spectral range for comparison. To distinguish spectral differences between healthy and infected samples, the random forest (RF) and K-Nearest Neighbors (KNN) machine learning algorithms have been employed. These algorithms automatically recognize spectral patterns associated with diabetes, enabling the prediction of unknown classifications based on established samples. Principal component analysis (PCA) was utilized for dimensionality reduction before feeding the data to RF and KNN for classification. The model’s classification performance was evaluated using 10-fold cross-validation, resulting in the proposed RF-based model achieving accuracy of 96 %, specificity of 100 %, sensitivity of 93 %, and precision of 100 %. These results suggest that the proposed method holds promise for a more convenient and potentially more accurate method for diagnosing glycosuria in diabetic patients.