Stress response caused by transportation is an important reason that affects the health and welfare of livestock, and the intelligent and precise monitoring of the stress state during transportation is a blank in research. On-site tracking experiments, literature review, and expert consultation indicate that environmental temperature, relative humidity and vibration intensity of vehicle are important stressors in the transportation process, and cortisol (COR) and adrenocorticotropic hormone (ACTH) are the key biomarkers to characterize the stress level. In this paper, stressors causing stress response were collected through environmental sensor network, and key biomarkers representing stress response were identified through biochemical analysis, and their change rules were analyzed and compared. Furthermore, the predictive models coupled with environmental parameters and stress markers were constructed through supervised learning networks. The results showed that: (1) After testing, the data collection performance of the multi-sensor network is reliable (P < 0.05); transportation will cause significant changes in the concentration of COR and ACTH (P < 0.05); the simulation results of stress markers data have no significant difference compared with the sample data (PSpline > PPchip > 0.90). (2) The validation results of models showed that: for COR, the absolute error and relative error based on general regression neural network (GRNN) are 2.59 ± 0.16 ng/mL and 1.30 ± 0.07%, respectively, which is better than the accuracy based on back propagation neural network (BPNN) and Elman neural network (Elman); for ACTH, the absolute error and relative error based on GRNN models were 1.75 ± 0.13 pg/mL and 2.14 ± 0.17%, respectively, which were better than the accuracy based on BPNN and Elman models. (3) For the prediction performance, the prediction accuracy of GRNN model for COR concentration is basically equal to that of Elman model, which is 0.17% higher than that of BPNN model for COR concentration, but the maximum relative error of BPNN model for COR concentration is the smallest. The prediction accuracy of Elman model for ACTH concentration is 0.09% higher than that of GRNN model, and 0.29% higher than that of BPNN model. But the maximum relative error of Elman model for ACTH concentration is the largest, and the running time is significantly longer. Although the performance of the three modeling methods is different, their overall accuracy has reached more than 97%. Therefore, this study can effectively identify the stress state of mutton sheep in the actual transportation process, and provide technical support for the health traceability and control of mutton sheep.