Further research directions
Section 6.1.2 demonstrated that many researchers applied numerical modeling approaches to study the heat and/or mass transfer through thermal protective clothing [35,36]. Similarly, Section 6.2.2 established that numerical modeling can also be developed to recognize metabolic heat and/or sweat-vapor transfer through clothing. These studies yield information about the mathematical interactions between fabrics’ attributes, thermal protective performance, and clothing comfort performance. However, practical applications of these mathematical models for calculating thermal protective performance or comfort performance may be limited due to their intrinsic complexities. To resolve these complexities, many researchers employed statistical techniques to easily understand the interactions between fabric attributes and thermal protective performance under different thermal exposures. By using statistical techniques, a set of multiple linear regression (MLR) models have been developed and validated to measure thermal protective performance [77,496]. However, the coefficients of determination (R2) and mean square error of these models are not so good; thus, the developed models may not accurately predict thermal protective performance. Many researchers have also challenged the inherent prediction accuracy of MLR models in different research fields (agriculture, applied chemistry, textile, etc.), especially where accurate forecasting is a prime concern [497-500]. They suggested that models developed by artificial neural network (ANN) could give better prediction accuracy than MLR models. According to them, ANN is an appropriate methodology for a wide range of applications through sufficient training; these applications can be especially helpful in building predictive models of processes where many issues (attributes of fabric, types of thermal exposures/intensity) contribute to the eventual outcome (thermal protective performance/clothing comfort), despite having little knowledge about the exact relationships or interactions between the input (attributes of fabric, types of thermal exposures/intensity) and output (thermal protective performance/clothing comfort) . The strength of ANN methodology lies in its ability to represent complex relationships, and learning about these relationships directly from the data being modeled . Due to this strength, Mandal and Song [77,496] recently applied the ANN method to predict thermal protective performance. However, they did not properly validate their developed models. It should be noted that while previous researchers mainly focused on the prediction of thermal protective performance using MLR or ANN, research on the prediction of clothing comfort is substantially limited.