AN EXPLAINABLE CNN-BASED APPROACH FOR EXOPLANET DETECTION USING LIGHT CURVE DATA

Abstract

Exoplanet detection has become one of the most significant research areas in modern astronomy due to its importance in understanding planetary systems and the possibil ity of extraterrestrial life. Among various detection techniques, the transit method is widely used because it identifies periodic dips in stellar brightness caused by a planet passing in front of its host star. The NASA Kepler mission has generated a massive amount of light curve data, making manual analysis difficult and time-consuming. Recent advancements in Artificial Intelligence and Deep Learning have enabled auto mated analysis of astronomical datasets. Convolutional Neural Networks (CNNs) have shown remarkable performance in extracting hidden patterns from light curve data and accurately classifying exoplanet signals. However, many deep learning models op erate as black-box systems and fail to provide interpretability, which limits trust and understanding in scientific applications. This thesis proposes an Explainable CNN-Based Approach for Exoplanet Detection using light curve data obtained from the NASA Kepler dataset. The proposed frame work focuses on both accurate prediction and interpretability of model decisions. A comparative analysis between CNN and ResNet-based architectures is also performed to evaluate classification performance in terms of accuracy, precision, recall, F1-score, and confusion matrix. Furthermore, explainable AI techniques are incorporated to vi sualize important light curve regions responsible for predictions, thereby improving transparency and reliability. Experimental results demonstrate that the proposed model achieves high detection ac curacy while also providing meaningful interpretation of predictions. The integration of explainable deep learning techniques with astronomical data analysis can support future space missions and improve the reliability of automated exoplanet discovery systems.