Methodical screening of safe and efficient drug candidate compounds is crucial for drug development. A high-throughput and accurate compound evaluation method targeting the central nervous system can be developed using in vitro neural networks. In particular, an evaluation system based on a human-derived neural network that can act as an alternative to animal experiments is desirable to avoid interspecific differences. A microelectrode array (MEA) is one such evaluation system, and can measure in vitro neural activity; however, studies on compound evaluation criteria and in vitro to in vivo extrapolation are scarce. In this study, we identified the parameters that can eliminate the effects of solvents from neural activity data obtained using MEA allow for accurate compound evaluation. Additionally, we resolved the issue associated with compound evaluation criteria during MEA using principal component analysis by considering the neuronal activity exceeding standard deviation (SD) of the solvent as indicator of seizurogenic potential. Overall, 10 seizurogenic compounds and three negative controls were assessed using MEA-based co-cultured human-induced pluripotent stem cell-derived neurons and astrocytes, and primary rat cortical neurons. In addition, we determined rat cerebrospinal fluid (CSF) concentrations during tremor and convulsion in response to exposure to test compounds. To characterize the in vitro to in vivo extrapolation and species differences, we compared the concentrations at which neuronal activity exceeding the SD range of the solvent was detectable using the MEA system and rat CSF concentration.