Published Paper Details:

Exoplanet research has grown into one of the most compelling frontiers in modern astrophysics reshaping what we know about worlds beyond our own solar system. These are planets orbiting stars other than the Sun and detecting them is rarely straightforward, astronomers typically rely on indirect methods, with transit photometry being among the most productive [1]. Space missions like the Kepler Space Telescope and TESS have generated enormous quantities of observational data, collectively responsible for confirming thousands of exoplanet candidates [3][4].Even so, a large share of that data has historically required manual review, which is both slow and inconsistent in practice. Classical detection tools like the Box Least Squares algorithm do a reasonable job of flagging periodic dips in light curves, but they struggle with noise and frequently flag non-exoplanets as real candidates. This paper introduces a hybrid detection framework that draws on traditional algorithms alongside machine learning and deep learning methods. Combining Random Forest classifiers, convolutional neural networks and oversampling via SMOTE, the system targets three key improvements: fewer false positives, stronger detection accuracy and the ability to scale across much larger observational archives.