Assembler: Efficient Discovery of Spatial Co-evolving Patterns in Massive Geo-sensory Data

Area of Science:

• Data Mining
• Geospatial Analysis
• Sensor Networks

Background:

• Geo-sensory applications are increasingly common, generating vast amounts of data from distributed sensors.
• Mining spatial co-evolving patterns (SCPs) is crucial for understanding correlated sensor behaviors but faces challenges.
• Challenges include filtering trivial fluctuations in time-series data and managing the large spatiotemporal search space.

Purpose of the Study:

• To develop an efficient and scalable method for mining spatial co-evolving patterns (SCPs) from massive geo-sensory data.
• To address the difficulties posed by noisy sensor readings and the combinatorial complexity of pattern discovery.

Main Methods:

• A two-stage approach named Assembler is proposed.
• Stage 1: Wavelet transform filters noise, and a segment-and-group method detects frequent individual sensor evolutions.
• Stage 2: Frequent evolutions are assembled into SCPs, utilizing a novel SCP search tree structure for efficient pruning.

Main Results:

• Assembler effectively filters trivial fluctuations and identifies frequent sensor evolutions.
• The SCP search tree enables efficient generation of spatial co-evolving patterns by pruning the search space.
• Experiments on real and synthetic datasets demonstrate Assembler's effectiveness, efficiency, and scalability.

Conclusions:

• Assembler provides an effective solution for mining spatial co-evolving patterns in large-scale geo-sensory data.
• The proposed method successfully handles noisy data and combinatorial complexity, offering a scalable approach.
• Assembler is a valuable tool for applications requiring the analysis of correlated sensor behaviors over space and time.