The 35th annual conference on Astronomical Data Analysis Software & Systems 2025 in Görlitz, Germany

This work analyzes machine learning approaches to exoplanet detection. We discuss methodological strengths and limitations including reliance on preprocessing, sensitivity to data imbalance, challenges with interpretability, and limited cross-mission generalization. We review more than 20 published machine learning (ML) models designed for this task including AstroNet, ExoNet, Genesis and ExoMiner. This review covers a number of ML architectures starting from deep learning techniques including convolutional neural networks (CNNs), deep learning networks (DNN), to the most recent transformers, as well as hybrid architectures. We explore different preprocessing workflows and data input formats such as phase-folded or raw light-curves, or pre-extracted parameters. We discuss the training datasets used and their impact on model performance and evaluation. We compare performance metrics such as accuracy, area-under-curve (AUC) and precision as they are reported in the original publications. Based on these criteria we summarize the current state of the art in ML for exoplanet detection by transit photometry.