Under Sampling Adaboosting Shapelet Transformation for Time Series Feature Extraction

摘要

To predict for machine defects, a classifier is required to classify the time series data collected from the sensors into the fault state and the normal state. In many cases, the data collected by sensors is time series data collected at various frequencies. Excessive computer load is required to handle this as it is. Therefore, there has been a lot of research being done on the process of extracting features that are highly classified from time series data. In particular, data generated at real-world is unbalanced and noisy, requiring time series classifiers to minimize their impact. Shapelet transformation is generally effectively known for classifying time series data. This paper proposes a process of feature extraction that is strong for noise and over-fitting to be applicable in practice. We can extract the feature from the time series data through the proposed algorithm and expect it to be used in various fields such as smart factory.