Machine learning pipeline for exoplanetary atmosphere characterization with Ariel Instruments.

This thesis looks into the problem of exoplanet atmospheric characterization through transit spectroscopy within the Ariel Data Challenge framework. The target is to retrieve the transmission spectra from the photometric time-series data but this becomes a problem as we have to deal with instrumental noises and limited spectral resolution. A machine learning assisted pipeline is developed that combines preprocessing with data driven models to estimate exoplanet transmission spectra from simulated ARIEL observation. Linear regression models with dimensionality reduction are first introduced to obtain baseline spectral estimates from high-precision photometric data. A convolutional neural network (CNN) is then applied as a non-linear refinement stage to improve spectral consistency across wavelength. The proposed approach demonstrates the effectiveness of hybrid physics-informed and machine learning methods for exoplanet atmospheric retrieval.