Mining of high utility-probability sequential patterns from uncertain databases.

Area of Science:

• Data Mining
• Pattern Recognition
• Uncertainty Quantification

Background:

• High-utility sequential pattern mining (HUSPM) is crucial for analyzing sequential data, with applications in consumer behavior and sensor networks.
• Existing HUSPM algorithms primarily focus on precise data, neglecting the inherent uncertainty in real-world data collection from sensors.
• Uncertainty, often represented by probabilities, is a significant factor in real-life databases, necessitating new mining approaches.

Purpose of the Study:

• To introduce a novel framework for mining high utility-probability sequential patterns (HUPSPs) in uncertain sequence databases.
• To address the limitations of existing methods in handling probabilistic data for high-utility pattern discovery.
• To develop an efficient algorithm capable of mining patterns under uncertainty.

Main Methods:

• A new framework, high utility-probability sequential pattern mining (HUPSPM), is proposed for uncertain sequence databases.
• A baseline algorithm incorporating three pruning strategies is presented for mining HUPSPs.
• A projection mechanism is employed to create smaller database projections, reducing candidate patterns and execution time.

Main Results:

• The proposed HUPSPM framework effectively mines HUPSPs in uncertain sequence databases.
• The algorithm demonstrates strong performance across real-life and synthetic datasets.
• Experimental results show improvements in runtime, candidate reduction, memory usage, and scalability.

Conclusions:

• The developed HUPSPM algorithm provides an efficient solution for discovering high-utility patterns in uncertain data.
• The projection mechanism significantly enhances the mining process by reducing computational overhead.
• The framework is adaptable to various minimum utility and probability thresholds, offering flexibility for diverse applications.