Improving the case-based reasoning prediction of the compliance of treated effluent from constructed wetlands

This study aims to improve a case-based reasoning system designed to predict the compliance for BOD5 of effluent from constructed wetlands in terms of simple-to-measure parameters. The data set was extended, similarity matching improved, the variable (field) weightings used in similarity scoring were refined using a genetic algorithm and the system was tested with an alternative CBR engine to verify the independence of the result. The modifications resulted in an improvement of overall accuracy from 81% to 83%, compared with 77% in a previous study, and an improvement of the accuracy of regulatory fail predictions from 58% to as high as 76%. Results were shown to be independent of the CBR engine used. Continuing inaccuracy is noted because the case-base includes many more pass than fail cases, and further improvements will be obtainable only by incorporating additional fail cases into the data.