Colorimetric analysis of saliva–alcohol test strips by smartphone-based instruments using machine-learning algorithms

Abstract

We report a smartphone-based colorimetric analysis of saliva–alcohol concentrations, utilizing optimal color space and machine-learning algorithms. Commercial saliva–alcohol kits are used as a model experiment, utilizing a custom-built optical attachment for the smartphone to obtain consistent imaging of the alcohol strips. The color of the strips varies with the alcohol concentration, and the smartphone camera captures the color produced on the test strip. To make a suitable library for each alcohol concentration, statistical methods were tested to maximize between-scatter and minimize within-scatter for each concentration. Results of three different classification methods (LDA, SVM, and ANN) and four-color spaces (RGB, HSV, YUV, and Lab) were evaluated with various machine-learning data sets and five different smartphone models. Cross-validation results were used to assess the statistical performance, such as positive (PPV) and negative (NPV) predictive values. An Android app developed and provided average classification rates of 100% and 80% for the standard and enhanced concentrations, respectively.