Multilevel sparse functional principal component analysis.

Area of Science:

• Statistics
• Data Science
• Functional Data Analysis

Background:

• Multilevel functional principal component analysis (MFPCA) is effective for densely sampled data.
• Sparse sampling presents challenges for traditional functional data analysis methods.
• Hierarchical data structures require specialized analytical techniques.

Purpose of the Study:

• To develop a method for analyzing sparsely sampled multilevel functional data.
• To adapt MFPCA for scenarios with limited observations per function.
• To enable accurate data reduction and curve reconstruction in sparse settings.

Main Methods:

• Exploiting the multilevel structure of covariance operators.
• Applying principal component decomposition at both between and within-subject levels.
• Developing methods for estimating covariance operators, functional principal component scores, and predicting underlying curves in sparse data.

Main Results:

• The proposed method effectively discovers dominant modes of variation in sparse multilevel functional data.
• Underlying curves are reconstructed accurately, even with few observations per function.
• Simulations demonstrate the method's robustness in sparse sampling contexts.

Conclusions:

• The developed approach provides a robust framework for analyzing sparsely sampled multilevel functional data.
• This method allows for effective data reduction and accurate curve prediction in challenging sparse data scenarios.
• The approach is applicable to diverse real-world datasets, including health studies and e-commerce data.