Pairing the positive and negative electrodes with their individual dynamic characteristics properly matched is essential to the optimal design of electrochemical energy storage devices. However, the complex relationship between the performance data measured for individual electrodes and the two-electrode cells used in practice often makes an optimal pairing experimentally challenging. In this work, taking graphene-based supercapacitors as an example, we combine experiments with machine learning to generate a large pool of capacitance data for graphene-based electrode materials with varied slit pore sizes and thicknesses, and numerically pair them into different combinations for two-electrode cells. The as-achieved pairing results allow us to conduct a comprehensive analysis of the correlations between the key electrode structural features of individual electrodes and volumetric capacitance of the resultant two-electrode cells. The results show that the optimal pairing parameters are varied considerably with the operation rate of the cells and are even influenced by the thickness of the inactive components. The best-performing individual electrode does not necessarily result in optimal cell-level performance. The machine learning-assisted pairing approach presents much higher efficiency compared with the traditional trial-and-error approach for the optimal design of supercapacitors and provides an additional effective avenue for further improving the performance of supercapacitors and is expected to play an enabling role in the future on-demand design of energy storage devices. The results observed in this work also indicate the call for comprehensive performance data reporting in the electrochemical energy storage field to enable the adoption of artificial intelligence techniques to accelerate the translation of academic research in this rapidly growing field.