Comparative Analysis of Machine Learning Models for Monitoring 4D Printed Smart Hydrogel Wearable Thermochromic Photonics Devices
Abstract
With the advancements in 3D printing, there is a growing interest in the 3D printing of polymer-based wearable sensors for monitoring environment parameters like temperature, UV, humidity, etc. For accurate measurement and prediction of these environmental parameters, linking such wearable sensors or their captured data with artificial intelligence is of interest. This study uses various machine learning regression models to predict the temperature of a 4D-printed smart hydrogel wearable that exhibits thermochromic properties. Red, green, and blue (RGB) values were obtained from images of the 4D-printed material samples in a temperature range of 25 – 50 °C. These RGB values (for over 1300 images) and their averages were used as the four predictor variables. Fifteen models were tested in total, consisting of simple linear regression (SLR), multiple linear regression (MLR), polynomial regression (PR), and multivariant polynomial regression (MPR), in which the predictors were taken in different combinations. The models were tested based on p-value, R2 score, root-mean-square deviation (RMSE) and normality and homoscedasticity of residuals. The MPR model of degree 3 with predictors blue and green performed the best (R2 = 0.8818, RMSE = 2.6390℃) and followed all the assumptions of polynomial regression.