Accelerated Mining of Partial Periodic Patterns in Temporal Datasets Using CUDA

Temporal databases play a crucial role in identifying patterns that offer insights into fields like fraud detection, market analysis, and healthcare by storing events in a sequential manner. Among these, partial periodic patterns are particularly valuable due to their ability to uncover behaviors that conventional frequent pattern mining techniques often overlook. These patterns allow for a relaxed strictness in the cyclic repetition of events, enabling the detection of patterns with missing occurrences. However, extracting partial periodic patterns, especially from large temporal datasets, is computationally intensive. Early research emphasized the need for advanced frameworks that can handle diverse periodic behaviors by introducing measures such as period support and techniques for managing periodic-frequent and infrequent patterns. This study presents a GPU-accelerated mining framework based on the state-of-the-art method 3P-BitVectorMiner, specifically designed for mining partial periodic patterns from temporal datasets. Leveraging CUDA's parallel processing capabilities, the parallel 3P-BitVectorMiner achieves significant improvements in speed and scalability. This work underscores the importance of GPU-accelerated approaches in enabling efficient and flexible analysis of partial periodic patterns in data-rich environments.