Exploring Misjudgments in IoT Analytics

In the industrial internet of things (IIoT), raw data are collected from thousands of sensors. Such a huge mass of data rises to the challenge of performing analytics. In the previous study, Pai et al. utilized varied methods to analyze the real-world paper breaks dataset. The preliminary test shows the difficulty in identifying anomalies, such as “non-linear relationship”, “normal instances with extreme value” and “identical patterns between normal and anomalous instances”. In this paper, for further study, we applied the non-linear support vector machines (SVM) method to classify anomalies. Given complete instances and class labels; however, the result of analytics is not 100% accuracy. In fact, there are 7 misjudgments. We considered that the class labels of raw data may be incorrect owing to certain error. According to PTS, the breaks should happen one after the other during a continuous period. However, in the dataset certain breaks are separate and their next instances are marked as normal operation. In sum, the problem of misjudgments is much more complicated. It's worth further studying from perspectives on both the quality of data and effectiveness of method.