Convolutional and Deep Learning based techniques for Time Series Ordinal Classification

TL;DR

The results obtained by ordinal versions of TSOC techniques are found to be significantly better than current nominal TSC techniques in terms of ordinal performance metrics, outlining the importance of considering the ordering of the labels when dealing with this kind of problems.

Abstract

Time Series Classification (TSC) covers the supervised learning problem where input data is provided in the form of series of values observed through repeated measurements over time, and whose objective is to predict the category to which they belong. When the class values are ordinal, classifiers that take this into account can perform better than nominal classifiers. Time Series Ordinal Classification (TSOC) is the field covering this gap, yet unexplored in the literature. There are a wide range of time series problems showing an ordered label structure, and TSC techniques that ignore the order relationship discard useful information. Hence, this paper presents a first benchmarking of TSOC methodologies, exploiting the ordering of the target labels to boost the performance of current TSC state-of-the-art. Both convolutional- and deep learning-based methodologies (among the best performing alternatives for nominal TSC) are adapted for TSOC. For the experiments, a selection of 29 ordinal problems from two well-known archives has been made. In this way, this paper contributes to the establishment of the state-of-the-art in TSOC. The results obtained by ordinal versions are found to be significantly better than current nominal TSC techniques in terms of ordinal performance metrics, outlining the importance of considering the ordering of the labels when dealing with this kind of problems.

Figures (9)

• Figure 1: Example of time series extracted from the DistalPhalanxOAG dataset. The target ordinal scale represents different age ranges.
• Figure 2: O-ResNet architecture. The notation x3 means that we have 3 stacked convolutional blocks inside each residual block. The final activation layer is the CLM (see \ref{['cap:clm_def']}).
• Figure 3: IM architecture. Two pipelines are applied in parallel to the input time series. On the first one, the bottleneck layer is applied, followed by $3$ sliding filter operations with sizes $40$, $20$ and $10$, respectively. On the second one, a Max Pooling is performed, followed by another bottleneck layer.
• Figure 4: O-IN architecture, two blocks of three IMs with residual connections are stacked, the resulting multivariate time series is passed to a GAP layer, followed by a fully connected and a CLM activation layer (see \ref{['cap:clm_def']}).
• Figure 5: Comparison between the TSOC methodologies and the baseline approaches described in \ref{['cap:proposed_methods']}. Methodologies are ordered based on the average rank over 30 resamples of train and test splits.