A R and Java implementation of a complete multivariate time series (MTS) outlier detection system covering problems from pre-processing to post-score analysis. It adopts dynamic Bayesian network (DBN) modeling with the conjunction of a sliding window approach to uncover observations suspected of not have been generated by data’s underlying processes. The system is formed by 4 main phases depicted in the image below. Most outlier detection methods focus solely on univariate series, providing analysts with strenuous solutions. The presented approach is capable of tackling multivariate longitudinal datasets through a widely accessible web application made available with a tutorial.
Source code and further instructions available here.