In vitro microelectrode array (MEA) assessment using human induced pluripotent stem cell (iPSC)-derived neurons holds promise as a method of seizure and toxicity evaluation. However it is difficult to detect the response of drugs with different mechanisms of action with a single parameter, and the analysis method has become an issue. Therefore, in this study, we developed an artificial intelligence (AI) model that learned raster plot images of electrical activity acquired by multiple electrodes and an SVM model that learned parameters calculated from time-series data of electrical activity. We compared the accuracy of predicting convulsive toxicity for each of the developed models. To train the model, extracellular potential data of co-cultured samples of human iPSC-derived cortical neurons and stellate cells obtained using MED64 Presto were used. In the SVM model using the spike time-series information and burst-related parameters, the risk assessment of the positive and negative compounds of the trained data was achieved, but the untrained data of acetaminophen was judged to be positive. In contrast, AI accurately predicted seizure risk, even with unlearned well data. These results indicated that using raster plot features method is useful for predicting the seizure liability using hiPSC-derived neurons.