Intensive livestock operations (e.g., in pig farming) generate significant environmental concerns due to emissions of both odorous (e.g., ammonia (NH3), hydrogen sulfide (H2S), Volatile organic compounds (VOCs)), and odorless (e.g., Methane (CH4), and carbon dioxide (CO2)) gases. These emissions impact animal welfare, productivity, and surrounding ecosystems. This paper develops and evaluates an Artificial Intelligence (AI) framework to forecast concentrations of carbon dioxide (CO2), a key indicator gas that serves as a proxy for the overall microbial and respiratory activity driving noxious emissions in pig farming environments. The system utilizes real-world data collected from gas sensors deployed at a pig farm in Vietnam. The data first undergoes comprehensive preprocessing, including cleaning, outlier detection, transformation, and normalization. A broad set of AI models is then implemented and evaluated, encompassing traditional statistical methods (e.g., ARIMA, Prophet), classical machine learning algorithms (e.g., Random Forest, XGBoost, SVM, KNN), and advanced deep learning architectures (e.g., LSTM, Transformer). The primary objective focuses on accurate CO2 concentration forecasting as a critical environmental indicator, which can indirectly contribute to the reduction of other odor-emitting gases. Experimental results show that modern machine learning and deep learning models achieve superior predictive performance compared to conventional approaches.