Research On Tensor Feature Selection Based On Genetic Algorithm

With the rapid development of computer and internet technology, web logs, image, audio,and video etc, which have been widely used have brought the sharply increase in amount ofdata, it indicates that the era of big data is coming. In the era of big data, the data hascharacteristics of the data amount increasing sharply and the more complicated data structureand data types. In the fields of pattern recognition, computer vision, and image processing,many real-world image and video data are more naturally represented as tensors. Withadvances in data collection and storage technologies, tensors are assuming increasingprominence in many applications and the problem of supervised tensor learning has emergedas a topic of critical significance in the research community. Recently, based on thesupervised tensor learning (STL) framework, a linear support higher-order tensor machine(SHTM) has been proposed.Considering that there are much redundancy information in the tensor data and the modelparameter largely affects the performance of SHTM, in this study, we present a geneticalgorithm (GA) based feature selection and parameter optimization algorithm for the linearSHTM, called TFS-SHTM. The proposed algorithm can remove the redundancy informationin tensor data and obtain better generalized accuracy by searching for the optimal modelparameter and feature subset simultaneously. A set of experiments is conducted on twelvetensor datasets to illustrate the performance of the proposed algorithm. The statistic test showsthat compared with the original linear SHTM, the proposed algorithm can provide significantperformance gain in term of generalized accuracy for tensor classification.Current research on supervised tensor learning mainly focuses on the study of lineartensor models. However, in real-world situations, data are often not linearly separable. Theconstruction of nonlinear tensor models becomes an important and challenging task. In thispaper, based on the fundamental principle of kernel method and the method of linear supporthigher-order tensor machine (SHTM), we develop a kernel-based tensor machine, calledNSHTM, which can be considered as an extension of nonlinear SVM to tensor patterns, but inner product computation operates on the decomposed data instead of its original form.Subsequently, considering that not all features are beneficial for the learning process andmodel parameters greatly affect the performance of the model, we introduce a geneticalgorithm (GA) to simultaneously perform tensor feature selection and parameteroptimization. A set of experiments is conducted on twelve tensor datasets to illustrate theperformance of the proposed algorithm. Extensive empirical studies show that NSHTM withgenetic algorithm can effectively improve tensor classification performances.