Moisture prediction in chicken litter using hyperspectral data and machine learning

• HSI provides a fast, non-invasive alternative to traditional moisture measurement. • XGBoost outperformed PLSR with raw data, achieving an R² of 0.97. • Preprocessing improved PLSR accuracy but reduced XGBoost performance. • Deep learning and data fusion can further enhance moisture prediction accuracy. Efficient moisture monitoring in chicken litter is essential for sustainable poultry waste management and its application as an organic fertilizer. Conventional methods like oven drying are destructive, time-consuming, and unsuitable for real-time assessment. While hyperspectral imaging (HSI) has been widely applied in agricultural sensing, its use for non-destructive moisture estimation in chicken litter remains largely unexplored. This study addresses that gap by evaluating the feasibility of combining HSI with machine learning models to predict moisture content in chicken litter. We employed two modeling approaches: Partial Least Squares Regression (PLSR) and Extreme Gradient Boosting (XGBoost), to assess performance using raw spectral data and under three preprocessing techniques designed to enhance data quality: Moving Average (MA), Multiplicative Scatter Correction (MSC), and Standard Normal Variate (SNV). XGBoost performed best on raw data (R² = 0.97, RMSE = 1.01), while PLSR improved notably with preprocessing, especially MA (R² = 0.92, RMSE = 1.56). These results reveal a key insight: preprocessing enhances linear model accuracy but may suppress critical non-linear patterns needed by tree-based models like XGBoost. Additionally, spatial visualization of moisture distribution across composting stages demonstrated the added value of HSI in characterizing heterogeneous organic materials. This study introduces a novel framework integrating HSI and machine learning for fast, non-destructive moisture assessment in poultry litter. It establishes a baseline for future work exploring deep learning, feature importance analysis, and multi-modal fusion to further improve accuracy and generalizability across diverse agricultural waste conditions.