Machine-Learned Decision Trees for Predicting Gold Nanorod Dimensions from Spectra Alone

Measuring the size of a single plasmonic nanoparticle is a critical step in correlating structure to optical properties but is often difficult to perform especially under irreversible conditions or non-immobilized nanostructures. Here, we demonstrate that gold nanorod dimensions can be accurately predicted over a wide range of aspect ratios using a simple machine learning model, decision trees. The model is trained using ~400 nanorod geometries and their corresponding scattering spectra obtained from finite-difference time-domain(FDTD) simulations. When validated on simulated FDTD spectra, 90 % of all predicted AuNR dimensions are within 10% relative error of the ground truth. Our results are tested using experimental spectra correlated with scanning electron microscopy images to obtain the sizes, estimating small and large nanorods dimensions with 8.5% relative error. Analysis of the decision tree structure reveals that a simple correlation with spectral peak position and linewidth of the localized surface plasmon resonance is adequate to predict nanorod dimensions, outperforming more complicated models. Our findings illustrate the advantages of using simple machine learning models to analyze single particle structural features from their optical spectra.