Comparative Analysis of XAI(eXplainable AI) for Optimization of Wastewater Treatment Process

In this paper, in order to optimize the biological water treatment process, we review three representative methods among XAI's post-hoc explainability techniques. Among them, LIME and AEA methods are applied to the water treatment biological process to find an optimization method. presented. XAI's post-hoc explainability technique is applied to solve the black box problem of not knowing what is attributable to the water treatment artificial intelligence model, which is commonly used in water treatment process optimization, even if it produces good results. We analyzed which control variables were responsible for the improvement in the quality of treated water. As a result of the analysis, it was confirmed that the LIME method had a greater influence on the quality of treated water than the AEA method. In addition, it was found that this method contributes to solving the black box problem and improving the quality of treated water. In the case of the LIME method applied to the water treatment biological process in this paper, although it is not common, it was possible to analyze the characteristics of output variables even for input variables other than control variables by observing the results graphically. In the future, it is believed that by systematizing such graph analysis into an algorithm, it will be possible to propose a more effective LIME method.