5th Place Solution: Light Curve Fitting Refined with Machine Learning

This solution involves a two-stage approach: (a) a main stage of physical model fitting to obtain a point estimate of the target parameter and (b) a refining stage to enhance accuracy. The light curve model calculates transit dips using numerical integration and accounts for additional effects such as limb darkening and orbital inclination. The model is fitted to noisy data using the Levenberg-Marquardt method, with the ingress / egress timings as optimizable parameters requiring only a rough initial guess. The refining stage removes noticeable deviations from the fitted planet radii, applies PCA and Gradient Boosting correction.