Abstract:In recent years, the methods on matrix-based or bilinear discriminant analysis (BLDA) have received much attention. Despite their advantages, it has been reported that the traditional vector-based regularized LDA (RLDA) is still quite competitive and could outperform BLDA on some benchmark datasets. Nevertheless, it is also noted that this finding is mainly limited to image data. In this paper, we propose regularized BLDA (RBLDA) and further explore the comparison between RLDA and RBLDA on another type of matrix data, namely multivariate time series (MTS). Unlike image data, MTS typically consists of multiple variables measured at different time points. Although many methods for MTS data classification exist within the literature, there is relatively little work in exploring the matrix data structure of MTS data. Moreover, the existing BLDA can not be performed when one of its within-class matrices is singular. To address the two problems, we propose RBLDA for MTS data classification, where each of the two within-class matrices is regularized via one parameter. We develop an efficient implementation of RBLDA and an efficient model selection algorithm with which the cross validation procedure for RBLDA can be performed efficiently. Experiments on a number of real MTS data sets are conducted to evaluate the proposed algorithm and compare RBLDA with several closely related methods, including RLDA and BLDA. The results reveal that RBLDA achieves the best overall recognition performance and the proposed model selection algorithm is efficient; Moreover, RBLDA can produce better visualization of MTS data than RLDA.

What problem does this paper attempt to address?

The main problems that this paper attempts to solve are some key challenges in multivariate time series (MTS) data classification. Specifically: 1. **Utilization of matrix data structure**: Existing multivariate time series classification methods usually do not fully utilize the matrix structure of data. Different from image data, the rows and columns of multivariate time series data represent different variables and time points respectively. Therefore, traditional linear discriminant analysis (LDA) and bilinear discriminant analysis (BLDA) methods may not be effective enough when dealing with this type of data. 2. **Singular matrix problem**: The existing BLDA method cannot be executed when one of the within - class matrices is a singular matrix. This limits the application of BLDA on certain datasets. 3. **Computational efficiency**: When the number of variables or time points is large, the computational cost of the existing BLDA method is high, which makes it less feasible in practical applications. To address these problems, the paper proposes a regularized bilinear discriminant analysis method (Regularized Bilinear Discriminant Analysis, RBLDA), and stabilizes the within - class matrix by introducing a regularization parameter, thus solving the singular matrix problem. In addition, the paper also develops an efficient implementation algorithm and a model selection algorithm to improve computational efficiency and classification performance. ### Main contributions 1. **Proposing RBLDA**: By regularizing the within - class matrix, RBLDA can handle the singular matrix problem in multivariate time series data, and improve computational efficiency while maintaining classification performance. 2. **Efficient implementation algorithm**: An efficient RBLDA implementation algorithm has been developed, which can quickly perform feature extraction and classification on large - scale datasets. 3. **Model selection algorithm**: An efficient model selection algorithm is proposed, which efficiently selects the regularization parameter through cross - validation, further optimizing the performance of RBLDA. 4. **Experimental verification**: Through experiments on multiple real - world datasets, the superior performance of RBLDA in multivariate time series classification tasks has been verified. ### Experimental results The paper conducted experiments on five publicly available real - world multivariate time series datasets, including the AUSLAN, ECG, Japanese Vowels, WAFER, and BCI datasets. The experimental results show that RBLDA achieves the best overall recognition performance on these datasets, and its visualization effect is also better than that of the traditional RLDA method. In conclusion, by proposing RBLDA and its efficient implementation and model selection algorithms, this paper successfully solves some key problems in multivariate time series data classification, providing new ideas and tools for research in this field.

Regularized Bilinear Discriminant Analysis for Multivariate Time Series Data

Pseudo Bidirectional Linear Discriminant Analysis For Multivariate Time Series Classification

A Report on Multilinear PCA Plus Multilinear LDA to Deal with Tensorial Data: Visual Classification As an Example

Two-Stage Regularized Linear Discriminant Analysis for 2-D Data.

Regularized Linear Discriminant Analysis Using a Nonlinear Covariance Matrix Estimator

Training Linear Discriminant Analysis in Linear Time

Linear Discriminant Regularized Regression

Regularized Deep Linear Discriminant Analysis

SRDA: an Efficient Algorithm for Large-Scale Discriminant Analysis

Block Linear Discriminant Analysis for Visual Tensor Objects with Frequency or Time Information

SRDA: an Efficient Algorithm for Large-Scale

Robust Bhattacharyya bound linear discriminant analysis through adaptive algorithm

Multilinear Class-Specific Discriminant Analysis

Multimodal Linear Discriminant Analysis Via Structural Sparsity

Robust sparse manifold discriminant analysis

Multivariate Functional Linear Discriminant Analysis for the Classification of Short Time Series with Missing Data

High-Dimensional Regularized Discriminant Analysis

Reverse nearest neighbors Bhattacharyya bound linear discriminant analysis for multimodal classification

Incremental Regularized Least Squares for Dimensionality Reduction of Large-Scale Data.

Linear Discriminant Analysis Using Rotational Invariant L1 Norm