Emissions of greenhouse gases are the primary cause of the global warming disaster. Much of the credit for the decrease in greenhouse gas emissions should go to the farming community. The world's governments are enacting policies to encourage the usage of low-carbon energy sources since their significance is increasing due to the worsening impacts of climate change. Sensors are finding applications in agriculture and food processing to control processes, track quality, and set standards for safety. Currently investigating machine learning models for CO2 emission prediction in the agri-food industry, as ML is increasingly being applied to address environmental challenges. By enhancing our capacity to foretell future energy demand, resource availability, and patterns and trends in low-carbon economic development, the scientific community and business stand to gain substantially from the use of artificial intelligence and big data analysis. This study introduces a hybrid model based on principles found in nature, called Bee-Lion. Its stated goal is to reduce agricultural and food-related carbon emissions. The feasibility of the methodology is demonstrated by the research's ability to efficiently predict carbon emissions and conduct low-carbon economic calculations using the hybridisation of two nature-based models, namely Artificial Bee Colony and ant-lion optimisation. By utilising the improved Bee-Lion algorithm, a subset of the agri-food carbon dioxide emissions dataset's most pertinent and superior features are chosen. Algorithms 7, 10, and 12 are chosen by the Bee-Lion, ABC, and ALO optimization algorithms, respectively, demonstrating their naturalistic inspiration. We can efficiently select the collection of features with the fewest possibilities using Bee-Lion. The random forest machine learning classifier helped us achieve a 98.78% accuracy rate.