COLOR ANALYSIS IN PAPER-BASED SENSING WITH IMAGE PROCESSING AND MACHINE LEARNING ANALYSIS FOR PREDICTING CHEMICAL PROPERTIES IN LIQUID SAMPLE

Abstract

Monitoring chemical properties in liquid samples, such as pH, glucose, and protein concentration, is essential in applications ranging from environmental water analysis to early-stage healthcare diagnostics. Multi-parameter test strips offer a commonly used and straightforward method for assessing these parameters in liquid samples, particularly in urine analysis. However, observer-dependent variability was noted in the interpretation of color changes, indicating potential bias among different evaluators. This study proposes a low-cost and accessible sensing method by integrating paper-based colorimetric sensing with image processing and supervised machine learning models to enhance the assessment using reagent test strips. Paper test strips were used as the medium for capturing chemical reactions., with color changes recorded using a PhotoBox equipped with a microcomputer and a Raspberry Pi camera. Extracted color features, such as Red (R), Green (G), Blue (B), Grayscale, Hue (H), Saturation (S), and Value (V) were used as inputs for predictive models, including Random Forest Regressor (RFR), Support Vector Regression (SVR), Gradient Boosting Regressor (GBR), Linear Regression (LR), K-Nearest Neighbors (KNN), and Multi-Layer Perceptron Neural Network (MLP-NN). Experimental results demonstrate that the proposed approach can accurately predict target values with strong predictive performance, achieving R² scores exceeding 0.90 in several models. Outlier handling and hyperparameter optimization were also conducted to improve prediction performance. These results suggest that this accessible and portable approach has significant potential for precise, scalable chemical assessment of multiple parameters (pH, glucose, and protein content) in liquid samples, supporting reliable monitoring in both environmental and healthcare-related applications.