Using Probabilistic Neural Networks to Analyze First Nations’ Drinking Water Advisory Data

Abstract

Drinking water advisories (DWAs) are a major issue facing many First Nations communities across Canada. This paper analyzes drinking water system data matched to DWA data using a probabilistic neural network (PNN) model to find key factors that influence the occurrence, frequency, duration, and cause of DWAs. First, for all data across Canada and subsequently for a number of data sets for individual provinces, the analyses were completed using an ensembles approach of running the same data set multiple times (in this case, five) to ensure that factors identified were representative of the entire data set and not just the training set. The results were compared with those from previous studies that used data mining techniques. Accuracies above 74% were achieved and key factors influencing each of the models were identified. The PNN models are a practical and powerful diagnostic tool for identifying key system attributes influencing DWAs, which can then be used to develop effective targeted solutions to mitigate the core issues that result in DWAs. These types of models can be used in other applications to identify influential factors and guide decision makers.