Benchmarking Unsupervised Clustering Algorithms for Atomic Force Microscopy Data on Polyhydroxyalkanoate Films.

Area of Science:

• Materials Science
• Polymer Science
• Data Science

Background:

• Polyhydroxyalkanoates (PHAs) are biodegradable polymers with diverse monomer compositions.
• Characterizing PHA film surfaces is crucial for understanding their properties and applications.
• Traditional surface analysis methods can be enhanced by advanced computational techniques.

Purpose of the Study:

• To classify polyhydroxyalkanoate (PHA) films based on surface characteristics using machine learning (ML).
• To benchmark the performance of 12 unsupervised clustering algorithms for PHA film analysis.
• To evaluate the impact of Fourier transform (FT) preprocessing on high-dimensional surface data.

Main Methods:

• Atomic Force Microscopy (AFM) was used to acquire surface topography data of PHA films.
• Unsupervised machine learning (ML) algorithms, including clustering, were applied to the AFM data.
• Vectorized surface data was preprocessed using 1D Fourier Transform (FT) for enhanced classification.

Main Results:

• The study benchmarked 12 widely used clustering algorithms for PHA film classification.
• Preprocessing vectorized data with a 1D Fourier Transform (FT) significantly improved classification accuracy.
• Specific algorithm performances varied depending on the dataset and data pool characteristics.

Conclusions:

• The 1D FT of vectorized AFM data is a highly effective method for classifying PHA films.
• Machine learning approaches offer powerful tools for advanced surface characterization of polymers.
• The developed tool provides a foundation for future improvements in automated surface investigation.