Explainable Galaxy Interaction Prediction with Hybrid Attention Mechanisms

TL;DR

This paper introduces an explainable hybrid neural network ensemble for galaxy interaction prediction, combining multiple architectures and interpretability tools to improve accuracy and reduce false positives in large-scale astronomical surveys.

Contribution

It presents a novel hybrid attention-based neural ensemble with explainability features, outperforming baseline models in galaxy interaction classification.

Findings

Achieved 96% accuracy and 0.97 F1 score.

Reduced false positives significantly compared to baseline.

Provided an interpretable framework suitable for large-scale surveys.

Abstract

Galaxy interaction classification remains challenging due to complex morphological patterns and the limited interpretability of deep learning models. We propose an attentive neural ensemble that combines AG-XCaps, H-SNN, and ResNet-GRU architectures, trained on the Galaxy Zoo DESI dataset and enhanced with LIME to enable explainable predictions. The model achieves Precision = 0.95, Recall = 1.00, F1 = 0.97, and Accuracy = 96%, outperforming a Random Forest baseline by significantly reducing false positives (23 vs. 70). This lightweight (0.45 MB) and scalable framework provides an interpretable and efficient solution for large-scale surveys such as Euclid and LSST, advancing data-driven studies of galaxy evolution.