Mining big spatial data

• Main Author: Ye, Ziyuan
• Other Authors: Gao Cong
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Computer science and engineering
• Online Access: https://hdl.handle.net/10356/166018
• Institution: Nanyang Technological University

The clustering of geospatial trajectories finds utility in multiple industries. In this report, two trajectory clustering models are presented, and a web-based front-end application is developed for visualization. The first trajectory clustering model contains two parts: i) trajectory reduction; ii) trajectory clustering. Trajectories are reduced using the Ramer–Douglas–Peucker algorithm and then clustered using Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN). The second trajectory clustering model adopts a recurrent neural network (RNN) model to learn spatial and temporal invariant representations of trajectories. It consists of three parts: (i) behavior feature extraction, (ii) trajectory representation learning, and (iii) trajectory clustering. For the behavior feature extraction, a sliding window technique is applied. For trajectory representation learning, a sequence-to-sequence autoencoder and decoder RNN network is trained to learn a fixed-length vector for each input trajectory. The vectors learned are then fed into the K-means clustering algorithm. Comparing the two models, we can conclude that trajectory pattern mining with sequence-to-sequence autoencoder and decoder RNN network yields more suitable results.