Exoplanet Detection Using Convolutional Neural Networks

Exoplanet surveys are creating large volumes of stellar photometric time series, making manual inspection of transit candidates costly and difficult to scale.This paper introduces a project-level framework for classifying stellar light curves with a onedimensional convolutional neural network (1D-CNN), at least empirically.The showed pipeline includes data loading, preprocessing, and normalization, supervised model training, validation, and candidate prediction, in actual application.The sketched network is using convolutional layers to learn local transit patterns, batch normalization to stabilize tuning, max pooling to diminish sequence length, dropout to limit overfitting, and thick layers to produce a binary grouping.Classic machine-learning baselines are also assessed by the report, covering Logistic Regression, Random Forest, and LightGBM, and motivates evaluating models with precision, recall, F1-score, ROC-AUC, and a confusion matrix rather than precision alone.The foundational project report outlines the CNN as reaching high classification accuracy, sometimes above 95%.Though, it doesn't offer the dataset split, class counts, confusion-matrix values, or a complete numerical results table needed to independently confirm that figure, so this paper is, broadly speaking, documenting the system design and evaluation approach without unverifiedly measuring as reproducible experimental results.