Performance Comparisons of Machine Learning Methods of PLA-based Photochromic Material UV Sensor

Abstract

A simple method for measuring of ultraviolet (UV) radiation or index values is introduced. In this study, which aims to use machine learning models to accurately analyze a changing color scale and make predictions about the magnitude of the external stimulus that causes color change, the photochromic Polylactic acid (PLA) material that changes color under UV light was video recorded with a smartphone camera. Then, by interpreting the data sets created from these images with machine learning models, a relationship was established between the current applied to the UV light source and the color. Video images taken with the smartphone camera were augmented with screen captures of 25 consecutive seconds, enabling the regression models used to make more accurate predictions. 9 different regression models were used, their performances were evaluated according to cross-validation results. Then, model performances were improved by using appropriate hyperparameters. Better accuracies were achieved especially in CatBoost Regression model. The findings of the study showed that UV intensity or index values can be determined with high accuracy with the existing smart phone camera without the need for any device.