Top-k Self-Adaptive Contrast Sequential Pattern Mining.

Area of Science:

• Computer Science
• Data Mining
• Machine Learning

Background:

• Sequential pattern mining (SPM) is crucial for identifying discriminative features in sequence classification.
• Existing contrast SPM methods suffer from complex parameter tuning and inefficient occurrence counting.
• Contrast pattern mining aims to find patterns with high contrast rates between categories for better interpretability.

Purpose of the Study:

• To propose a top-k self-adaptive contrast SPM algorithm to address limitations of existing methods.
• To enhance sequence classification accuracy and pattern interpretability.
• To develop an efficient method for discovering contrast patterns.

Main Methods:

• Introduced a top-k self-adaptive contrast SPM (SCP) approach that adaptively adjusts gap constraints.
• Utilized a Nettree data structure with special arrays for efficient pattern occurrence counting.
• Developed Zero and Less pruning strategies and a Contrast-first mining strategy to handle non-Apriori properties.

Main Results:

• The proposed algorithm demonstrates significant efficiency in mining contrast patterns.
• Contrast patterns derived from the algorithm show superior performance in sequence classification tasks compared to frequent patterns.
• Experimental validation confirms the effectiveness of the self-adaptive contrast SPM method.

Conclusions:

• The self-adaptive contrast SPM method offers an efficient and effective approach for feature discovery in sequence classification.
• Contrast patterns are more discriminative than frequent patterns for classification tasks.
• The Nettree structure and proposed mining strategies improve computational efficiency.