Machine learning for small data sets: an exemplary study on the classification of highly complex surface micromorphologies
Source abstract
Abstract
Complex micro-morphologies can dictate the physical behavior of materials and their interactions with other objects but are difficult to classify. Here, machine learning may be a useful tool; however, missing expertise and small samples sizes often prevent experimental scientists from making use of this technique. Here, we present an exemplary study on how to apply machine learning to a topography-based problem. We analyze highly complex surface wear patterns on samples subjected to single damages and dual damage combinations. We demonstrate how human bias can be avoided when class definitions are ambiguous ( e.g ., when experimental groups exhibit strong class overlaps), i.e. , by employing a principal component analysis in combination with a clustering algorithm. Furthermore, we describe how 7 distinct classes of surface topographies can be accurately classified: we compare the performance of a rather simple random forest model with that of a custom-made, yet easy to (re)use convolutional neural network (CNN), and we assess the influence of low data availability. Finally, we describe how pre-training of CNNs can be used to analyze very small or imbalanced datasets. This work equips researchers with a model pipeline of procedures to perform a complex surface analysis by combining unsupervised and supervised learning.
Source: openalex_inverted_index · Quality: baseline
Research context
OpenAlex topics: Topological and Geometric Data Analysis · Morphological variations and asymmetry · Theoretical and Computational Physics
SciVX discovery tags are terminology-based. OpenAlex topics are source-provided classifications.
References (42)
Show source-provided references
1. Impact of geometry on chemical analysis exemplified for photoelectron spectroscopy of black silicon10.1002/smtd.202401929 ↗
2. Emergence of self-affine surfaces during adhesive wear10.1038/s41467-019-09127-8 ↗
3. Measuring the complexity of micro and nanostructured surfaces10.1016/j.matpr.2021.10.120 ↗
4. Contact between rough surfaces and a criterion for macroscopic adhesion10.1073/pnas.1320846111 ↗
5. Exploration of the nanomedicine-design space with high-throughput screening and machine learning10.1038/s41551-019-0351-1 ↗
6. Machine learning assisted hit prioritization for high throughput screening in drug discovery
7. Machine learning-driven data valuation for optimizing high-throughput screening pipelines10.1021/acs.jcim.4c01547 ↗
8. High-throughput screening and general synthesis strategy of single-atom nanozymes for oral squamous cell carcinoma therapy10.1002/adma.202416463 ↗
9. Direct microplate sampling mass spectrometry for high-throughput screening of biocatalytic activity10.1021/acs.analchem.5c00267 ↗
10. Machine learning-assisted high-throughput screening of nanozymes for ulcerative colitis10.1002/adma.202417536 ↗
11. Machine learning-driven high-throughput screening of alloy-based catalysts for selective CO2 hydrogenation to methanol10.1021/acsami.1c16696 ↗
12. Discovery of surface-engineered nanoparticles that boost enzyme activity via high-throughput screening and machine learning
13. Small data machine learning in materials science10.1038/s41524-023-01000-z ↗
14. Rapid and flexible segmentation of electron microscopy data using few-shot machine learning10.1038/s41524-021-00652-z ↗
15. Data augmentation in microscopic images for material data mining10.1038/s41524-020-00392-6 ↗
16. Logic replicant: a new machine learning algorithm for multiclass classification in small datasets
17. A strategy to apply machine learning to small datasets in materials science10.1038/s41524-018-0081-z ↗
18. A virtual sample generation algorithm supporting machine learning with a small-sample dataset: a case study for rubber materials10.1016/j.commatsci.2022.111475 ↗
19. Using deep neural network with small dataset to predict material defects10.1016/j.matdes.2018.11.060 ↗
20. Intelligent prediction of wear location and mechanism using image identification based on improved faster R-CNN model10.1016/j.triboint.2022.107466 ↗
21. Wear mechanism classification using artificial intelligence10.3390/ma15072358 ↗
22. Boosting tree-assisted multitask deep learning for small scientific datasets10.1021/acs.jcim.9b01184 ↗
23. Mechanism-guided wear severity assessment of worn surfaces with multiple damages10.1016/j.wear.2025.205875 ↗
24. Employing knowledge transfer in machine learning for wear assessment on synthetic and biological materials
25. Suppressed droplet splashing on positively skewed surfaces for high-efficiency evaporation cooling10.1002/smll.202307759 ↗
26. Switching between 3D surface topographies in liquid crystal elastomer coatings using two-step imprint lithography10.1002/smll.202302051 ↗
27. Surface topology affects wetting behavior of Bacillus subtilis biofilms10.1038/s41522-017-0018-1 ↗
28. Cell morphology as a design parameter in the bioengineering of cell–biomaterial surface interactions10.1039/D1BM01149A ↗
29. Cellular response to surface morphology: electrospinning and computational modeling10.3389/fbioe.2018.00155 ↗
30. Biological responses to immobilized microscale and nanoscale surface topographies10.1016/j.pharmthera.2017.07.009 ↗
31. Surface roughness gradients reveal topography-specific mechanosensitive responses in human mesenchymal stem cells10.1002/smll.201905422 ↗
32. Effect of surface morphologies on the In vitro and In vivo properties of biomedical metallic materials10.1021/acsbiomaterials.4c00942 ↗
33. Unsupervised machine learning to analyze corneal tissue surfaces10.1063/5.0159502 ↗
34. In vitro method to quantify and visualize mechanical wear in human meniscus subjected to joint loading10.1016/j.jmbbm.2022.105338 ↗
35. Time wears on: assessing how bone wears using 3D surface texture analysis10.1371/journal.pone.0206078 ↗
36. Synthesis and characterization of a lubricin mimic (mLub) to reduce friction and adhesion on the articular cartilage surface10.1016/j.biomaterials.2015.09.012 ↗
37. Imaging structural and electrical changes of aging cells using scanning ion Conductance microscopy10.1002/smtd.202301315 ↗
38. NanoLocz: image analysis platform for AFM, high-speed AFM, and localization AFM
39. On the class overlap problem in imbalanced data classification10.1016/j.knosys.2020.106631 ↗
40. K. He, X. Zhang, S. Ren, and J. Sun, "Deep residual learning for image recognition," in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 27–30 June 2016 2016, pp. 770–778, doi: 10.1109/CVPR.2016.90.10.1109/CVPR.2016.90 ↗
41. An efficiency-driven, correlation-based feature elimination strategy for small datasets10.1063/5.0118207 ↗
42. Henkel · GitHub · 2026