To examine brain areas related to the cognitive load condition during the Stroop task, we proposed a method using a Deep Neural Network (DNN). We acquired cerebral blood flow data in congruent and incongruent tasks by near-infrared spectroscopy (NIRS) equipped with 22 ch. The data were used to train a DNN, and the influence of each factor on the output was evaluated. Our DNN model consists of independent input layers for each channel of NIRS, as well as fully-connected hidden layers and output layers. Our results suggest that the medial prefrontal cortex (focusing on cognition) and the left inferior frontal gyrus (focusing on language processing) were involved in the cognitive load during the Stroop task. These results in the Stroop task were consistent. Therefore, the proposed method’s utility was confirmed.