Constructing compact Takagi-Sugeno rule systems: identification of complex interactions in epidemiological data

Area of Science:

• Data Mining
• Computational Biology
• Epidemiology

Background:

• Takagi-Sugeno (TS) fuzzy rule systems are valuable for identifying non-linear variable interactions.
• High-dimensional data presents challenges due to the 'curse of dimensionality,' leading to an explosion in the number of rules.
• Existing methods lack robust approaches for rule selection, limiting applications in complex fields like epidemiology and bioinformatics.

Purpose of the Study:

• To develop a parsimonious Takagi-Sugeno (TS) fuzzy rule system for high-dimensional data.
• To introduce novel statistics (R, L, and ω-values) for ranking TS rule importance.
• To implement a forward selection procedure for constructing parsimonious TS models.

Main Methods:

• Development of a new parsimonious TS rule system.
• Proposal of R, L, and ω-values for ranking TS rule importance.
• Application of a forward selection procedure for model construction.

Main Results:

• A parsimonious TS model was successfully constructed using a subset of rules.
• The model accurately describes the relationship between deprivation indices and educational outcomes.
• Selected rules highlight synergistic relationships and context-dependent effects of deprivation domains.

Conclusions:

• The developed parsimonious TS rule system effectively handles high-dimensional data and the curse of dimensionality.
• The findings underscore the importance of considering synergistic interactions between variables in complex systems.
• This method offers a valuable tool for identifying non-linear interactions in biomedical and epidemiological data, informing policy decisions.