Machine Learning Prediction of Litter Moisture and Litter pH in a Commercial Broiler House Using Nested Leave-One Production-Period-Out Cross-Validation

Litter moisture content (LMC) and litter pH (LPH) are important indicators of litter quality in broiler production. This study developed machine-learning models to predict LMC and LPH using environmental, flock-related, spatial, and litter-related variables collected from eight temporally distinct production periods in the same commercial broiler house at days 7, 21, and 42 (1600 observations). Four ML algorithms, namely Support Vector Regression (SVR), Random Forest (RF), Extreme Gradient Boosting (XGB), and Multilayer Perceptron (MLP), were evaluated under six predictor scenarios (S1–S6) using nested Leave-One-Production-Period-Out cross-validation. For LMC, XGB under S3 achieved the highest mean outer-fold R2 (0.796), XGB under S5 the lowest RMSE (3.023) and MAPE (8.356%), and SVR under the litter-assisted S6 scenario the lowest MAE (2.433). For LPH, RF showed consistently competitive performance and achieved R2 = 0.891, RMSE = 0.181, MAE = 0.139, and MAPE = 1.922% under the litter-assisted S6 scenario. These comparisons were interpreted as exploratory, metric-specific benchmarking rather than definitive model selection. The nested framework provided grouped internal validation across held-out production periods within the studied house. Because all periods originated from one house, broader generalization requires validation in additional commercial facilities. Prospective validation is also required before operational deployment or assessment of management, welfare, environmental, or sustainability benefits.