Abstract
Understanding the long-term evolution of hierarchical triple systems ischallenging due to its inherent chaotic nature, and it requires computationallyexpensive simulations. Here we propose a convolutional neural network model topredict the stability of hierarchical triples by looking at their evolutionduring the first $5 \times 10^5$ inner binary orbits. We employ the regularizedfew-body code \textsc{tsunami} to simulate $5\times 10^6$ hierarchical triples,from which we generate a large training and test dataset. We develop twelvedifferent network configurations that use different combinations of thetriples' orbital elements and compare their performances. Our best model uses 6time-series, namely, the semimajor axes ratio, the inner and outereccentricities, the mutual inclination and the arguments of pericenter. Thismodel achieves an area under the curve of over $95\%$ and informs of therelevant parameters to study triple systems stability. All trained models aremade publicly available, allowing to predict the stability of hierarchicaltriple systems $200$ times faster than pure $N$-body methods.