With sensors percolating through everyday living, it may be toted that there is an enormous increase in the availability of real-time streaming and time series data. We also see an exponential increase in number of industry applications with sensors driven by IoT and connected with data sources that change over time. This time-series data presents many technical challenges, opportunities, and threats to industries. Thus, streaming analytics to model an unsupervised machine learning system for detecting unusual/anomalous behavior in real-time must be prominently addressed. In this paper, the authors propose a real-time abnormality detection model using a Facebook prophet that addresses issues related to the improper Facebook collection of data, further leading to faulty analysis and wrong results. The proposed unsupervised model detects abnormalities in the data captured through customer order by considering day and date as constraints. The proposed model is found to be even more efficient in RMSE score. The proposed model delivered enhanced performance compared to other traditional approaches.
展开▼
