A novel approach for incremental uncertainty rule generation from databases with missing values handling: application...

Area of Science:

• Data Mining
• Medical Informatics
• Database Systems

Background:

• Traditional association rule mining assumes static databases, which is unsuitable for dynamic medical data with missing values.
• Home care systems and other medical databases often contain missing attribute values, posing challenges for standard data mining techniques.

Purpose of the Study:

• To introduce URG-2 (Uncertainty Rule Generator), a novel algorithm designed for mining association rules in dynamic databases with missing values.
• To evaluate the performance of URG-2 against classical approaches in handling missing data in medical datasets.

Main Methods:

• URG-2 utilizes a single pass over the dataset to generate item sets, employing new metrics for Support and Confidence.
• The algorithm was tested on two medical databases with simulated missing values (5-20% rate).
• Performance was compared to a classical approach that ignores records with missing values.

Main Results:

• URG-2 demonstrated greater robustness in mining rules from datasets with missing values.
• Rule preservation was 30-60% higher with URG-2 compared to the classical approach.
• The incremental nature of URG-2 resulted in over 90% time savings for re-mining.

Conclusions:

• URG-2 offers a robust and efficient solution for association rule mining in dynamic relational databases, particularly in medical contexts with missing data.
• The algorithm effectively handles missing values, preserving more rules and significantly reducing computational time.
• URG-2 presents a preferable alternative for data mining applications where data completeness is not guaranteed.