A parallel varied density-based clustering algorithm with optimized data partition

Lyu

et al. 2022

Transactions in GIS

Self Cite

Most existing trajectory clustering methods use point matching similarity measurement strategies to compare trajectory objects; however, such methods have poor similarity measurement accuracy, which reduces clustering quality. To address this research gap, this article considers a combination of trajectory‐feature learning and clustering. A deep trajectory clustering (DTC) method toward clustering‐friendly spaces is proposed to simultaneously learn trajectory representations (i.e., low‐dimensional vectors are used to represent trajectories) and cluster centroids, and further acquire the cluster assignments end to end. We first use a sequence‐to‐sequence autoencoder network for pretraining to obtain preliminary trajectory vectors, and then jointly optimize the learned representations and cluster centroids under the same framework. The effectiveness of DTC is demonstrated in synthetic datasets, and the learned representations in the real world can be used to identify hot routes in cities.

Section: Related Workmentioning

confidence: 99%

A deep trajectory clustering method based on sequence‐to‐sequence autoencoder model

Lyu

et al. 2022

Transactions in GIS

Self Cite

“…DBSCAN [11] defines a cluster with points from continuous highdensity regions and treats those points in low-density regions as outliers or noises. Inspired by this popular algorithm, a lot of density-based clustering methods have been designed, such as OPTICS [2], DENCLUE [15], DESCRY [1], and others [10,12,21,24]. DenPeak (clustering by fast search and find of density peaks ) [29] is another immensely popular density-based clustering method, which assumes that cluster centers locate in regions with higher density and the distances among different centers should be relatively large.…”

Section: Density-based Clusteringmentioning

confidence: 99%

Deep density-based image clustering

Ren

Knowledge-Based Systems

et al. 2020

Recently, deep clustering, which is able to perform feature learning that favors clustering tasks via deep neural networks, has achieved remarkable performance in image clustering applications. However, the existing deep clustering algorithms generally need the number of clusters in advance, which is usually unknown in real-world tasks. In addition, the initial cluster centers in the learned feature space are generated by k-means. This only works well on spherical clusters and probably leads to unstable clustering results. In this paper, we propose a two-stage deep densitybased image clustering (DDC) framework to address these issues. The first stage is to train a deep convolutional autoencoder (CAE) to extract low-dimensional feature representations from high-dimensional image data, and then apply t-SNE to further reduce the data to a 2-dimensional space favoring density-based clustering algorithms. The second stage is to apply the developed density-based clustering technique on the 2-dimensional embedded data to automatically recognize an appropriate number of clusters with arbitrary shapes. Concretely, a number of local clusters are generated to capture the local structures of clusters, and then are merged via their density relationship to form the final clustering result. Experiments demonstrate that the proposed DDC achieves comparable or even better clustering performance than state-of-the-art deep clustering methods, even though the number of clusters is not given.

“…The STME algorithm is time consuming during the search for k-neighbours, and future research may focus on reducing its runtime. Some data partitioning strategies have been proposed to reduce the search time for k-neighbours (Gu et al, 2017), and many popular distributed computing frameworks, such as Hadoop and Twister, have also been applied to spatiotemporal clustering algorithms for improving efficiency (Ghuli, Shukla, Kiran, Jason, & Shettar, 2015), which are all directions for us to research in the future. Furthermore, the STME algorithm can be applied to other geographical data sets to help identify multi-type clustering patterns, such as spatiotemporal correlation analysis of different types of infectious diseases, forecast of traffic rush hours and places, and evaluation of the distribution rationality of public facilities.…”

Section: Con Clus I On S and Future S Tud Ie Smentioning

confidence: 99%

“…A spatiotemporal event can be labelled as a spatiotemporal point with location, time stamp and type. Most spatiotemporal event clustering algorithms were designed for spatiotemporal events of the same type (Brás Silva, Brito, & Joaquim, 2006;Cheam, Marbac, & McNicholas, 2017;Gu et al, 2017). However, there are many random phenomena involving spatiotemporal events with multiple types in real life, such as origins and destinations of taxi trips, multiple points of interest (POI), multi-covariate patterns of vegetation succession, and moving in and out of migratory processes (Gao, Kupfer, Zhu, & Guo, 2016;Tao & Thill, 2019).…”

Section: Introductionmentioning

confidence: 99%

STME: An effective method for discovering spatiotemporal multi‐type clusters containing events with different densities

et al. 2020

Transactions in GIS

Self Cite

Clustering on spatiotemporal point events with multiple types is an important step for exploratory data mining and can help us reveal the correlation of event types. In this article, we present an effective method for discovering spatiotemporal multi‐type clusters containing events with different densities and event types (STME). Particularly, the type of events in a cluster can be different, and clusters with similar densities but different internal compositions should be distinguished. We use the distance to the kth nearest neighbour to define the size of the searched neighbourhood, and expand clusters by the concept of cluster reachable, ensuring that the proportion of various types of events in the cluster remains stable. The concept of clustering priority is also proposed to make the cluster always expand from the region with the highest density, which improves the robustness of clustering. Moreover, the density of multiple types of events in clusters is estimated to discover the internal structure of clusters and further explore the correlation between events. The effectiveness of the STME algorithm is demonstrated in several simulated and real data sets, including points of interest data in Beijing and the origins and destinations of taxi trips in New York.