Recently, ammonia decomposition has attracted attention as an eco-friendly hydrogen production and storage technology because it can generate hydrogen without carbon emission. However, catalysts are essential to promote ammonia decomposition reactions. To date, ruthenium(Ru) is widely known as an ammonia decomposition catalyst showing high activity. However, ruthenium is a precious metal and is expensive, making it less economical to commercialize it as a catalyst. Therefore, more and more attempts are being made to explore multi-component alloy catalysts mixed with cheap elements to find catalysts to replace ruthenium. However, it takes a very long time to explore the candidate group of multi-component alloy catalysts through replicates and first principle calculations, which are traditional search methods. In this work, we propose a method to explore and design alloy catalysts to replace ruthenium using the machine learning model slab graph convolutional neural network(SGCNN) to accelerate catalyst search. At this time, through the machine learning model used, it was possible to quickly predict the adsorption energy that can infer catalytic activity by entering only the adsorption structure for the reactants, intermediate products, and products of the ammonia decomposition reaction. In addition, by learning a machine learning model for the collected single and heterogeneous alloy catalyst materials, it was possible to quickly and accurately predict the ammonia decomposition catalyst activity value for the multi-component alloy. Finally, a new catalyst consisting of inexpensive elements with catalytic performance similar to that of ruthenium was explored.