A novel framework for time series clustering based on distribution and geometric distance
Smit Thakkar · Queensland University of Technology · 2023
Recognizing patterns in transport data is crucial for planning and traffic management. This research proposes a framework for clustering transport time series data by modeling objects as probability distributions. It introduces a robust distance measure, utilizes Kernel Density Estimation for distribution distances, incorporates a composite measure for temporal trends, and offers intuitive cluster evaluation. Applied to travel time data from Finucane Road in Brisbane, it demonstrates effectiveness and broader applicability. By explicitly considering probability distributions, this framework addresses limitations of traditional divergence measures. It provides a systematic approach for handling large transport-related data, facilitating informed decision-making in various transport applications.