This paper aims to construct the interest prediction models for nursery school child using a single-channel electroencephalograph (EEG). Recently, the number of dual income households who leave their children in nursery schools have been increasing in Japan. Such parents are not able to grasp their children's behavior in daily life. Considering these issues, the researches related to child behavioral analysis have been proceeded by using image data taken from digital cameras. However, it is difficult to acquire the behavioral information from the digital cameras at anytime, anywhere. Therefore, we are focusing on wearable systems for keeping an eye on a child. Specifically, we adopt the EEG to design the constructing system. In this paper, we acquire single-channel EEG recordings from nursery school children when they watch picture-story shows. Furthermore, we apply a non-negative matrix factorization (NMF) to artifactitious rejection and a genetic algorithm-partial least squares (GA-PLS) regression to detect important frequency components and design the interest prediction models for the child using a single-channel EEG. As a result, we showed that over 60% estimation accuracy could be obtained all except one subject and the specific combinations of the frequency components selected by the GA-PLS, and we also could confirm that the NMF could remove the eye blink artifacts.