TCN-XGBoost Hybrid Model for Daily FCR Prediction and One Step Forecasting in Broiler Farm

Abstract

This study presents a hybrid TCN-XGBoost model for predicting daily Feed Conversion Ratio (FCR) in broiler chicken farming, addressing critical gaps in numerical FCR forecasting using sequential data. Feed expenses constitute approximately 70% of total operating costs in broiler production, yet many semi-modern farms in Indonesia still rely on manual, reactive FCR calculations. While existing research has focused on binary classification or species other than broilers, this study specifically targets numerical FCR prediction by integrating Temporal Convolutional Networks (TCN) for temporal feature extraction with Extreme Gradient Boosting (XGBoost) for robust regression. The research utilized 15 production cycles from Misjiwati Farm in North Sumatra, encompassing daily metrics of feed intake, body weight, mortality, and FCR. A comprehensive feature engineering pipeline was developed, incorporating lagged features, rolling window statistics, momentum metrics, and interaction features to capture both short-term and long-term dependencies. Hyperparameter optimization using Optuna resulted in optimal configurations: sequence length of 11 days, batch size of 64, TCN dropout rate of 0.4, and XGBoost with 775 estimators. The model demonstrated exceptional predictive performance with R² = 0.9532, MAE = 0.0131, RMSE = 0.0169, and MAPE = 1.05%, significantly exceeding thresholds for excellent biological system predictions. Single-step forecasting validation achieved 0.426% relative error, confirming practical deployment viability. Residual analysis revealed homoscedastic behavior with a near-zero mean residual (0.006093) and tight standard deviation (0.015786), validating statistical reliability across all FCR ranges. The model successfully predicted 92% of values within ±2 standard deviations, with only 8.3% exhibiting residuals exceeding ±0.035. This hybrid architecture establishes a scalable solution for precision poultry farming, enabling proactive feed management interventions and early warning systems for performance deterioration, offering significant potential for enhancing profitability and sustainability in Indonesian broiler chicken production.