A novel machine learning-based method for identifying Industrial Symbiosis opportunities

Industrial Symbiosis (IS) is a sub-field of Industrial Ecology (IE) that aims to reuse waste streams as resources among industries to achieve higher resource efficiency. A common method for identifying IS opportunities is input–output matching, which links the output (waste) of one company to the input of another. In such a process, obtaining data on resource and waste streams of participating companies becomes a prerequisite. However, data acquisition is challenging due to issues like data confidentiality and the demand of extensive manual input. It is known that process information is generally kept confidential, resulting in lack of data on types of wastes and resources available in the system. Moreover, collecting such data, typically via industrial workshops, is time-consuming and costly. To address the problem, we propose to take a predictive approach to obtain the input and output streams of companies using advanced Machine Learning (ML) techniques, eliminating the need for companies to register their process data prior to IS implementation. Specifically, we develop a novel Neural Network (NN) model to predict whether a waste is an input or output of an activity based on the textual descriptions of the waste and the activity. The developed models achieve an F1-score higher than 0.82 on the test set. We further validate the model on an external unseen dataset and demonstrate its application in identifying IS opportunities through a case study. By predicting the input and output data, the proposed method enables input–output matching without extensive information sharing by companies, serving as a preliminary step for identifying IS opportunities automatically.