Machine learning method for predicting line-shapes of Fano resonances induced by bound states in the continuum

We consider resonances induced by symmetry protected bound states in the continuum in dielectric gratings with in-plane mirror symmetry. It is shown that the shape of the resonance in transmittance is controlled by two parameters in a generic formula which can be derived in the framework of the coupled mode theory. It is numerically demonstrated that the formula encompasses various line-shapes including asymmetric Fano, Lorentzian, and anti-Lorentzian resonances. It is confirmed that the transmittance zeros are always present even in the absence up-down symmetry. At the same time reflectance zeros are not generally present in the single mode approximation. It is found that the line-shapes of Fano resonances can be predicted to a good accuracy by the random forest machine learning method which outperforms the standard least square methods approximation in error by an order of magnitude in error with the training dataset size [Formula: see text].