MIFuzzy Clustering for Incomplete Longitudinal Data in Smart Health

Area of Science:

• Data Science
• Machine Learning
• Health Informatics

Background:

• Missing data frequently occur in longitudinal smart health studies, impacting data analysis.
• Multiple imputation (MI) is a standard technique for handling missing data but its application in unsupervised learning needs further investigation.
• Fuzzy clustering is a valuable tool for analyzing complex datasets, including those with missing values.

Purpose of the Study:

• To introduce and evaluate the MIFuzzy clustering approach for handling missing data in smart health studies.
• To theoretically, empirically, and numerically demonstrate the performance of MI-based fuzzy clustering.
• To compare the uncertainty of clustering accuracy between MI-based and traditional imputation methods.

Main Methods:

• Developed and applied the MIFuzzy clustering algorithm, an extension of previous MIfuzzy clustering work.
• Utilized non-parametric soft computing techniques for semi-supervised and unsupervised learning.
• Compared MIFuzzy clustering against non-imputation and single-imputation based clustering approaches.

Main Results:

• MI-based fuzzy clustering significantly reduces uncertainty in clustering accuracy compared to other methods.
• The MIFuzzy approach demonstrates robust performance in processing incomplete longitudinal behavioral intervention data.
• Theoretical, empirical, and numerical analyses confirm the utility and strength of the proposed method.

Conclusions:

• MIFuzzy clustering offers a powerful solution for analyzing incomplete longitudinal data in smart health.
• This approach enhances the reliability of clustering results, crucial for understanding treatment effect heterogeneity.
• The study advances the application of imputation techniques within unsupervised learning frameworks for health research.