Support tensor data description
- Journal of Quality Technology
- April 2021
- Volume 53 Issue 2
- pp. 109-134
- Maboudou-Tchao, Edgard M.
Many problems in pattern/image recognition, fault diagnosis, signal processing, anomaly detection, and machine learning generate massive amounts of multidimensional data with multiple aspects and high dimensionality. Multi-way arrays or tensors provide a natural and compact representation for such massive multidimensional data. Big data analytics require new technologies to efficiently deal with huge datasets within acceptable elapsed times. In some applications, we encounter a one-class classification problem due to availability of data of only one class. Recent advances in technology allow collecting data in high dimensional structure, namely tensors. Traditional vector-based algorithms have limitations when dealing with tensor input data. In this work, a new one-class classification method to deal with tensors as original input, support tensor data description (STDD), is introduced. This classifier works on tensor space. The advantage of this new proposal over one-class support tensor machine (OCSTM) is that support tensor data description does not use the alternating projection method of one-class support tensor machine. Consequently, support tensor data description is easier and faster to implement than one-class support tensor machine. The efficiency of the proposed method over one-class support tensor machine in some cases is illustrated through simulations. Also, this proposal is compared with a vector-based method and the experimental results illustrate the efficiency of the tensor-based algorithm over the vector-based one.