Objective: Seizure occurs when the balance between excitatory and inhibitory (E/I) inputs to neurons is perturbed, resulting in abnormal electrical activity. This study investigated whether an existing E/I imbalance in neural networks is a useful diagnostic biomarker for Rolandic epilepsy by a resting-state dynamic causal modeling-based support vector machine (rs-DCM-SVM) algorithm. Methods: This multicenter study enrolled a discovery cohort (76 children with Rolandic epilepsy and 76 normal controls [NCs]) and a replication cohort (59 children with Rolandic epilepsy and 60 NCs). Spatial independent component analysis was used to seven canonical neural networks, and a total of 25 regions of interest were selected from these networks. The rs-DCM-SVM classifier was used for individual classification, consensus feature selection, and feature ranking. Results: The rs-DCM-SVM classifier showed that the E/I imbalance in brain networks is a useful neuroimaging biomarker for Rolandic epilepsy, with an accuracy of 88.2% and 81.5% and an area under curve of.92 and.83 in the discovery and the replication cohorts, respectively. Consensus brain regions with the highest contributions to the classification were located within the epilepsy-related networks, indicating that this classifier was suitable. Consensus functional connection pairs with the highest contributions to the classification were associated with an excitation network loop and an inhibition network loop. The excitation loop mediated the integration of advanced cognitive networks (subcortex, dorsal attention, default mode, executive control, and salience networks), whereas the inhibition loop was involved in the segregation of sensorimotor and language networks. The two loops showed functional segregation. Significance: Brain E/I imbalance has potential to serve as a biomarker for individual classification in children with Rolandic epilepsy, and might be an important mechanism for causing seizures and cognitive impairment in children with Rolandic epilepsy.