TSFuse: automated feature construction for multiple time series data

A central paradigm for building predictive models from time series data is to convert the into a feature vector representation and then apply standard inductive learners. Typically, conversion done by manually defining features, which an extremely time-consuming error-prone process. This has motivated development of algorithms that automatically construct features series. However, these systems are typically designed univariate data. In contrast, many real-world applications require analyzing consisting collected multiple sensors. this context, it often useful derive new fusing both within sensor across different Unfortunately, poses additional challenges automated construction as exponentially more operations possible than in case. paper proposes system called TSFuse, supports fusion explores search space computationally efficient way. We perform empirical evaluation on classification datasets show our able find better compared existing