Ethanol poses a significant threat to driver safety, as its effects vary with blood alcohol concentration (BAC). Common methods for estimating BAC include breath alcohol analysis, which calculates BAC from the alcohol concentration in exhaled breath, and direct blood sampling. However, these methods have notable limitations. This study aims to classify alcohol concentration using non-invasive optical signal data obtained from biomimetic samples with varying alcohol levels. To replicate the high scattering characteristics of biological tissue, scattering effects were induced in the samples, and absorbance was measured using near-infrared (NIR) wavelengths, which penetrate biological tissue more deeply. A Random Forest (RF) model was trained using the measured absorbance values to classify alcohol concentration levels. The Area Under the ROC Curve (AUC) for each concentration level indicated effective model learning, and the classification results on the test set demonstrated statistically significant accuracy. These findings suggest that the RF model can classify alcohol concentrations non-invasively and without the loss of samples. Furthermore, incorporating additional optical properties beyond absorbance may improve the accuracy of future non-invasive alcohol concentration classification models.

Alcohol acts as a central nervous system depressant and is classified as a psychoactive drug that impairs cognitive alertness and motor coordination. Driving after alcohol consumption slows reaction time in emergency situations and increases the risk of collisions. Although various technologies have been developed to measure alcohol concentration, many suffer from limitations such as sensitivity to environmental factors (e.g., temperature and humidity), hygiene concerns, and the need for periodic calibration. This study proposes an optical method for measuring alcohol concentration using near-infrared (NIR) spectroscopy. Statistical analyses were conducted across multiple wavelength regions to identify wavelengths with significant correlations to alcohol concentration. As a preliminary step, single alcohol solution samples were prepared using distilled water and ethanol. The optical properties of the samples were analyzed in the NIR wavelength range, and statistical indicators including the coefficient of determination (R²), p-value, and coefficient of variation (CV) were evaluated. The results identified specific wavelengths with statistical significance, and the application of mathematical modeling at these wavelengths enabled accurate estimation of alcohol concentration. This approach demonstrates the potential for non-invasive alcohol concentration measurement while minimizing hygiene and infection-related concerns for users.