Spatially Encoding Temporal Correlations to Classify Temporal Data Using Convolutional Neural Networks

TL;DR

This paper introduces a method that transforms temporal data into images using Gramian Angular Fields and Markov Transition Fields, enabling CNNs to classify time series effectively by leveraging computer vision techniques.

Contribution

It presents a novel approach to encode temporal correlations spatially as images, allowing CNNs to learn from time series data, which is a new application of vision-based deep learning in this domain.

Findings

Competitive classification accuracy on benchmark datasets

Effective feature learning from GAF and MTF images

Insights into CNN feature importance for temporal data

Abstract

We propose an off-line approach to explicitly encode temporal patterns spatially as different types of images, namely, Gramian Angular Fields and Markov Transition Fields. This enables the use of techniques from computer vision for feature learning and classification. We used Tiled Convolutional Neural Networks to learn high-level features from individual GAF, MTF, and GAF-MTF images on 12 benchmark time series datasets and two real spatial-temporal trajectory datasets. The classification results of our approach are competitive with state-of-the-art approaches on both types of data. An analysis of the features and weights learned by the CNNs explains why the approach works.