Time-Series Data Analysis for Classification of Noisy and Incomplete Internet-of-Things Datasets

Detecting and classifying device types and their long term communication patterns and anomalies in massive, noisy and anonymized Internet-of-things (IoT) data sets is a challenging problem. Recent advances in computational approaches for Topological Data Analysis (TDA), including the technique of persistence homology, appear to offer tremendous possibles for understanding highly complex IoT data sets. This paper presents the results of our use of TDA to understand a data set captured over 9 months of hundreds of interacting IoT devices situated in multiple residential settings. The data set is noisy, incomplete and subject to multiple Pattern-of-Life (PoL) fluctuations. We treated the data set as a collection of multi-attribute time series and performed several types of IoT classification experiments. We compared our results to other single and multi-attribute techniques for time series analysis. The outcome was that, as compared to these other standard methods, TDA does particularly well for classifying incomplete, noisy and PoL dependent IoT data.