Akademik Veri Yönetim Sistemi
2nd International Cumhuriyet Artificial Intelligence Applications Conference, Sivas, Türkiye, 8 - 09 Aralık 2022, ss.26-30, (Tam Metin Bildiri)
The Internet of Things (IoT) enables the
emergence of huge amounts of time series data. Hence, the
classification/clustering of time series data is a hot research
field in data mining. Deep embedded clustering (DEC) is a
novel unsupervised learning algorithm that concurrently
devises the feature representations and cluster assignments. In
this study, the performance of DEC on IoT time series data is
investigated. Therefore, two variants of DEC are considered
for the performance assessment on IoT time series data. The
first deep clustering algorithm is the original DEC which uses
k-means in the clustering layer while the second deep
clustering algorithm is DEC using the mean shift method in the
clustering layer. In the experiments, nine datasets from the
UCR time series classification archive are used. The
experimental results reveal that DEC using the mean shift
shows better clustering performance than DEC using k-means
on IoT time series data when there are enough data samples in
the dataset to set apart the clusters. Otherwise, DEC using kmeans will be a better choice to perform clustering tasks on
IoT time series data.