Ensemble Based Positive Unlabeled Learning for Time Series Classification

摘要: Many real-world applications in time series classification fall into the class of positive and unlabeled (PU) learning. Furthermore, many these applications, not only are negative examples absent, available for learning can also be rather limited. As such, several PU algorithms have recently been developed to learn from a small set P labeled seed augmented with U examples. The key is accurately identify likely U, but it has remained challenge, especially those uncertain located near boundary. This paper presents novel ensemble based approach that restarts detection phase times probabilistically label more robustly so reliable classifier built limited training Experimental results on data different domains demonstrate new method outperforms existing state-of-the art methods significantly.