XAI-SFD: an explainable artificial intelligence framework for intelligent digitalisation of shit flow diagrams and smart sanitation management in South Africa

Despite the availability of safe sanitation access, this issue remains a major challenge in many developing nations like South Africa where there are challenges associated with the fragmented data gathering, manual reporting, and lack of interoperability and real-time monitoring of the sanitation service chain. Shit Flow Diagrams (SFDs) are among the tools that can be used for the presentation and analysis of the sanitation service chain as well as identification of potential risks in the sanitation systems; however, the creation of traditional SFDs is a manual process and leaves no space for continuous monitoring. This paper proposes XAI-SFD, an Explainable Artificial Intelligence (XAI)-based solution for the intelligent digitalization of Shit Flow Diagrams (SFDs) and smart sanitation management in South Africa. This approach is based on IoT data gathering, GIS-based spatial analysis, data validation, machine learning, explainable artificial intelligence, sanitation service chain modelling, Digital Twin technology and dashboard decision-making support. Explainability provides transparency and allows the stakeholders to understand what influences the risk assessment and prioritization. A proof-of-concept was carried out on a controlled municipality scale, using 247 monitoring stations of the Routine Inland Water Quality Monitoring network of the City of Cape Town. A total of four clusters of spatial contexts were identified, with the selected clustering model providing a silhouette coefficient of 0.3133 and Davies-Bouldin index of 1.3324. As a result of the explained priority assessment, 25 monitoring stations (10.12%) have been defined as belonging to the high-priority class, 99 stations (40.08%) belonging to the medium-priority class, and 123 stations (49.80%) belonging to the low-priority class. Because of the lack of sanitation flows outcomes and water quality monitoring results in the dataset on monitoring sites, it is possible to consider the results received as prototyping findings and not as the validation of sanitation risk prediction models. In this regard, the proposed XAI-SFD architecture may be used as the base for further creation of heterogeneous databases of sanitation and environmental data.