A compression of artificial neural network and regression models for prediction of dry matter digestibility in some tropical forages

Dry matter digestibility (DMD) of forages is very important in ruminant nutrition. Because in vivo and in vitro procedures to determine dry matter digestibility are time-consuming and expensive, providing an easy and fast technique would be of interest. Modeling approach is a novel procedure for predicting an output from some predictors in a system. In this study, DMD of several tropical forages were predicted from chemical analyses including crude protein (CP), crude fiber (CF), ash and acid detergent fiber (ADF). Five tropical forages were harvested in 10 replicates and each sample was analyzed for CP, CF, ash and ADF as inputs and DMD as output. A total of 50 data lines were divided into training (35 data lines) and testing (15 data lines) subsets. Three modeling approaches including artificial neural network (ANN), principal component analysis (PCA), and partial least squares (PLS) were implemented; and performance of each model was evaluated using coefficient of determination (R2), mean square error (MSE), mean absolute deviation (MAD), mean absolute percentage error (MAPE), and bias. The results showed that ANN-based model outperformed the classical regression methods. The ANN-based model with the highest R2 (0.95) and lowest residual distribution (MSE = 0.871, MAD = 0.772, and MAPE = 1.44) gave the best predictions for DMD.