Multivariate LSTM-FCNs for Time Series Classification

TL;DR

This paper introduces multivariate extensions of LSTM-FCN and ALSTM-FCN models with squeeze-and-excitation blocks, achieving superior accuracy and efficiency in complex time series classification tasks with minimal preprocessing.

Contribution

The paper presents novel multivariate LSTM-FCN and ALSTM-FCN models incorporating squeeze-and-excitation blocks, enhancing accuracy and efficiency over existing models.

Findings

Outperform most state-of-the-art models in multivariate time series classification.

Require minimal preprocessing for effective performance.

Efficient at test time and suitable for deployment on memory-constrained systems.

Abstract

Over the past decade, multivariate time series classification has received great attention. We propose transforming the existing univariate time series classification models, the Long Short Term Memory Fully Convolutional Network (LSTM-FCN) and Attention LSTM-FCN (ALSTM-FCN), into a multivariate time series classification model by augmenting the fully convolutional block with a squeeze-and-excitation block to further improve accuracy. Our proposed models outperform most state-of-the-art models while requiring minimum preprocessing. The proposed models work efficiently on various complex multivariate time series classification tasks such as activity recognition or action recognition. Furthermore, the proposed models are highly efficient at test time and small enough to deploy on memory constrained systems.