Smartphone-Based Colorimetric Analysis of pH Strips Using Machine Learning

Abstract

This study introduces a machine learning (ML)-enhanced smartphone application designed for the precise colorimetric quantification of pH strips. To ensure the robustness of the system against environmental variations, a comprehensive dataset was constructed by capturing images of pH strips under diverse illumination conditions and camera angles. Following region of interest extraction, an initial set of 33 colorimetric features was employed to train and evaluate 15 different regression models. To ensure model interpretability and computational efficiency, a SHapley Additive exPlanations (SHAP)-based analysis was implemented, successfully identifying six critical descriptors (including color channel skewness, entropy, and intensity metrics) that primarily govern the pH prediction. The best-performing model (R2 = 0.99) was subsequently integrated into a user-friendly Android application, pHScoper. This application enables image capture, interactive cropping, and offline, on-device quantitative analysis without cloud reliance. Overall, the developed platform demonstrates strong potential for reliable, low-cost pH measurements in resource-limited settings.