Abstract—Due to its various applications, time-series classifi-
cation is a prominent research topic in data mining and computational
intelligence. The simple k-NN classifier using dynamic
time warping (DTW) distance had been shown to be competitive
to other state-of-the art time-series classifiers. In our research,
however, we observed that a single fixed choice for the number of
nearest neighbors k may lead to suboptimal performance. This
is due to the complexity of time-series data, especially because
the characteristic of the data may vary from region to region.
Therefore, local adaptations of the classification algorithm is
required. In order to address this problem in a principled way by,
in this paper we introduce individual quality (IQ) estimation. This
refers to estimating the expected classification accuracy for each
time series and each k individually. Based on the IQ estimations
we combine the classification results of several k-NN classifiers
as final prediction. In our framework of IQ, we develop two
time-series classification algorithms, IQ-MAX and IQ-WV. In our
experiments on 35 commonly used benchmark data sets, we show
that both IQ-MAX and IQ-WV outperform two baselines.
Index Terms—time series; classification; individual quality (IQ)
I. INTRODUCTION
Classification of time series is a prominent research topic
in data mining and computational intelligence, because of
its numerous applications in various domains such as speech
recognition [1], signature verification [2], brainwave analysis
[3], handwriting recognition, finance, medicine, biometrics,
chemistry, astronomy, robotics, networking and industry [4].
The simple 1-nearest neighbor (1-NN) method using dynamic
time warping (DTW) distance [1] has been shown to
be competitive or superior to many state-of-the art time-series
classification methods [5], [6], [7]. However, the choice of
parameter k in the k-NN classifier is known to affect the
bias-variance trade-off [8]: smaller values of k may lead to
overfitting, whereas larger values of k increase the bias and
in this case the model may capture only global tendencies.
Recent studies [9] have indicated that significant improvement
in the accuracy of the k-NN time-series classification can be
attained with k being larger than 1. This is due to intrinsic
characteristics in time-series data sets, such as the mixture
between the different classes, the dimensionality, and the
skewness in the distribution of error (i.e., the existence of “bad
hubs” [9] that account for a surprisingly large fraction of the
total error). Parameter k can be chosen using a hold-out subset
of the training data.