This study aimed to explore the application of the support vector regression (SVR) algorithm that assisted Raman spectroscopy in detecting metabolites in athletes' urine. With the development of sports science, metabolite detection of athletes has become an important means of evaluating their training effects and physical state. Metabolites such as urea and creatinine in urine are essential indicators for monitoring the metabolic function of the human kidney. In this study, artificial urine samples with different urea and creatinine contents were detected by self-developed near-infrared micro confocal Raman spectroscopy, and the spectral differences between urine samples were analyzed. The vibrational spectra of urea and creatinine molecules were predicted based on density functional theory. The Raman characteristic peak intensities of urea and creatinine and their concentrations were analyzed and compared, and the support vector regression models with different kernel functions were established. The optimal quantitative analysis model predicted the contents of urea and creatinine in human urine. The urea and creatinine determination coefficients were 0.9815 and 0.9681, respectively, and the root mean square errors were 0.0345 and 1.4591, respectively. Based on the quantitative analysis model constructed by metabolites, 800 urine samples of 40 athletes were quantitatively detected for urea and creatinine. The comparison between the detection results and the traditional test results showed that this study's method had high clinical application potential. In a word, near-infrared micro confocal Raman spectroscopy technology is a rapid, convenient, no-pretreatment, and no-damage detection analysis method with great application potential. Its repeatability and stability make it suitable for rapidly screening kidney diseases in people with special diseases.