Application of artificial intelligence in the hazard indexes of recycled agricultural waste materials

Radioactive substances are emitted during the recycling of agricultural waste materials, putting both the environment and people at risk. Thus, the research forecasts the risks from these materials, using deep neural networks (DNN) with a variety of network architectures. Levenberg–Marquardt backpropagation was used as a training algorithm and the neural network was built using just one target variable, the hazard index, together with three input variables consisting of 226Ra, 232Th, and 40 K. The model was trained using 3–5-5–5-1, 3–10-10–10-1, and 3–15-15–15-1 network architectures. Additional datasets were used to validate the developed model. The results showed that every agricultural byproduct evaluated provides no potential indoor and outdoor risks. All network structures yielded strong precision for predicting the hazard indexes of agricultural byproducts. However, when compared to alternative network topologies, a 3–10-10–10-1 network architecture showed the best performance metrics for training, validation, and testing. In addition, the confirmation of the model with untrained data yielded a strong correlation with 98.68