Abstract: Abstract: Environmental quality of swine gestation barns can bring a significant impact on the fertility of breeding sows. Therefore, it is crucial to accurately evaluate the environmental quality and then timely trim the conditions, particularly for high breeding efficiency under less environmental stress. In this study, an environmental quality evaluation model of swine gestation barns was proposed using the Simulated Annealing-Particle Swarm Optimization-Least Absolute Shrinkage and Selection Operator-Back Propagation (SA-PSO-LASSO-BP) neural network (NN). Firstly, nine parameters were identified using the Chinese National Criteria. A data collection system was then established to collect the environmental data. Secondly, a Kalman filter and a batch estimation adaptive weighted fusion algorithm were introduced to fuse the multi-node environmental data, in order to remove the errors and redundant data from the data collection. Thirdly, a least absolute shrinkage and selection operator (LASSO) regression model was selected for the feature selection. There were four feature factors that were closely related to environmental quality, including temperature, relative humidity, NH3 concentration, and CO2 concentration. Meanwhile, the structural parameters were optimized in the BP-NN , where the number of hidden layer nodes was determined to be 11. Finally, the initial weights and threshold values of the BP NN were optimized by the SA-PSO for the ultimate evaluation model. A comparison was made on the several NNs to verify the evaluation performance of the SA-PSO-LASSO-BP NN, including the BP, LASSO-BP, Genetic Algorithm-LASSO-BP (GA-LASSO-BP), Sparrow Search Algorithm-LASSO-BP (SSA-LASSO-BP), and the PSO-LASSO-BP NN. The training results proved that the convergence accuracy and rate of the SA-PSO-LASSO-BP network were significantly improved by the feature selection with the LASSO regression model. The number of iterations and the network errors of the LASSO-BP NN decreased by 29.9% and 17.78%, respectively, compared with the BP NN. In terms of feature selection, the SA-PSO algorithm implemented by the SA-PSO-LASSO-BP NN was utilized to optimize the initial weights and thresholds of the network, in order to further improve the convergence accuracy and rate of the model. Compared with the PSO-LASSO-BP NN, the number of iterations, running time, and network error were reduced by 54.43%, 17.24%, and 8.33%, respectively. The validation test indicated that the best performance of the model was achieved, with the coefficient of determination (R2) of 0.918, an overall accuracy of 95.85%, the Mean Absolute Error (MAE) of 0.037, and the Root Mean Squared Error (RMSE) of 0.176. Consequently, the SA-PSO-LASSO-BP NN model can better fit the nonlinear relationship between complex environmental factors and environmental quality. The finding can serve as a strong reference for the environmental quality evaluation of swine gestation barns.