Predicting the effect of fabric properties on the thermal insulation of sportswear by using artificial neural network and multiple linear regression approaches

The study looks at two machine learning methods-multiple linear regression and artificial neural networks-to predict the thermal insulation value, known as CLO, of knit sportswear fabrics. The artificial neural network is created using feed-forward backpropagation and the trainlm training function in MATLAB. Its weights and basic values are adjusted using the Levenberg-Marquardt optimization method. The network uses the sigmoid transfer function to set the layer output and track its performance. When it comes to accuracy and how well it handles complex situations, the artificial neural network outperforms the multiple linear regression model. It better captures how thermal insulation values relate to fabric qualities. For the artificial neural network model, the mean absolute error percentage was 2.2771, the root mean squared error was 0.0017, and the coefficient of determination was 0.9268. The multiple linear regression model had a coefficient of determination of 0.8046 and mean absolute error percentages of 4.4170 and 0.0027. Compared to a time-consuming trial-and-error approach, the study shows that artificial neural networks are a better way to predict thermal insulation in textiles. This emphasizes how important these models are for creating energy-efficient designs and improving material engineering.