Dimensionality Reduction Method of Multivariate Time Series Based on Classification Projection

The feature information of minority samples is often overlooked when dealing with imbalanced data sets in multivariate time series using traditional methods for dimensionality reduction. In light of this issue, a classification projection-based method for dimensionality reduction in multivariate time series is proposed in this paper. Firstly, the dataset is classified using prior label information and class covariance information is extracted for each class. Subsequently, an average covariance matrix is computed from these individual class covariance matrices, which undergoes singular value decomposition to obtain the projection matrix. Ultimately, the goal of reducing data dimensionality is achieved. Experimental results demonstrate that when confronted with imbalanced data sets in multivariate time series, the proposed approach enhances classification accuracy and reduces computational complexity compared to traditional methods.