Odor concentration estimation using the principal component regression ; in the grow to finish swine pen

Abstract

In this study, a odor concentration estimation model in the grow to finish swine pen was derived using principal component analysis and multiple regression models. 14 kinds of odor compounds were used. Prior to the statistical analysis, the missing value (less than MDL)for each compounds was imputed with MDL/2, and the concentration of the odor compounds and the odor unit were log-transformed to follow a normal distribution. The three principal components extracted through principal component analysis explained for 77.03% of the initial variables. The correlation between Principal components and odor compounds was evaluated by the loading value. PC1 was correlated with VOC (10 Type), PC2 with sulfur compounds (3 Type), and PC3 with ammonia and hydrogen sulfide. A odor concentration estimation model was derived using PC1, PC2, PC3 as independent variables. As a result, the model and the regression coefficient were statistically significant (p < 0.05), but adjusted R 2 (0.25) was low.