Deep Learning for Extracting Human Movement Patterns from Spatio-Temporal Data

Human mobility data play an important role across various domains and applications by providing valuable insights into the movement of individuals, groups, and populations. Mining these data improves the ability to extract simple and complex human mobility patterns. These patterns are crucial for understanding the human behavior and predicting next-location of human movements that are essential for many applications, such as traffic forecasting, epidemic spreading, urban planning, and commuting flows. Great research efforts are spent to understand complex characteristics reflected in human mobility, and challenges related to trajectory data, such as time sensitivity, privacy risk assessment, sparsity, and higher-order dependencies. In this paper, we review recent approaches for extracting human movement patterns using deep learning techniques. We discuss these techniques and show their relevance to the different dimensions of human mobility data, challenges of trajectory data, and applications they can serve. Through this study, we reveal the limitations of traditional machine learning models, and how deep learning models can overcome these limitations. Additionally, we discuss the important factors of privacy risk assessment that should be considered in human mobility. Finally, our review helps researchers to understand challenges of human mobility and demonstrate the future directions for leveraging deep learning to extract human movement patterns using the power of graph-based methods.