Utilizing discrete choice models for fusing alarms from multiple water quality indicators

Current attempts for event detection utilize a variety of approaches ranging from statistical models, to heuristics, machine learning and optimization methods. In this study, we combine the different approaches to build a high performance event detection system. A common feature with most developed event detection systems (EDSs) is that alarms are obtained as per water quality indicators. Unifying the single alarms from the different indicators was considered by means of simple heuristics. A salient feature of the developed approach is using a statistically oriented model for discrete choice prediction which is estimated using a maximum likelihood method. Specifically, each of the water quality indicators is considered as an expert with its own opinion regarding contamination event probability. A linear utility function which uses the experts' probabilities as input variables is then used to inform the discrete choice model on the selection between event and non-event decisions. The discrete choice model is jointly calibrated with other components of the EDS framework in a training data using genetic algorithms. As such, the proposed framework saves us from the need for heuristics to unify single indicator probabilities which was proven to be a crucial part in other EDSs.