In copolymerization, the monomers' reactivity ratios play an important role in shaping the final copolymer properties. Thus, the knowledge of reactivity ratio is essential to the polymer reaction engineering. However, the experimental methods of determining reactivity ratios are laborious and require many repetitions and testing at different compositions. Therefore, computational methods for determining the reactivity ratios based on the chemical structures of the reactants have attracted significant attention. Here, we demonstrate how machine learning enables the prediction of comonomers reactivity ratios based on the molecular structure of monomers. We combined multi-task learning, multi-inputs, and one of the state-of-the-art machine learning model – Graph Attention Network – to build a model capable of predicting reactivity ratios based on the monomers’ chemical structures. We found that the interpretable characteristics of Multi-Input-Multi-Output Graph Attention Network, along with its capability to learn chemical features facilitates the accurate prediction of the reactivity ratios. Such a predictive tool can be used in combination with macroscopic kinetic models to design new copolymers.