Time receiver operating characteristic (tROC) curves: A new tool for evaluating the performance of diagnostic systems

Rapid and reliable fault detection is critical for the safe and economic operation of complex industrial systems like aircraft and locomotives. A metric for quantifying the performance of fault detection systems is important for selecting and optimizing such systems. Traditional two-class classifier metrics such as accuracy or ROC curves are inappropriate for time series data. This paper introduces the tROC, a method of quantifying performance of fault detection systems that has many of the desirable properties of traditional ROC curves, while accounting for the time series nature of the data.