This study aims to analyze the developmental characteristics of early childhood musical expressions from a viewpoint of movement elements, and to devise a method to evaluate the development regarding musical expression in early childhood using machine learning. Previous studies regarding motion capture have shown analysis results such as specific actions and responses to music (Burger et al, 2013). In this study, firstly, ANOVA was attempted on full-body movements. The author quantitatively analyzed the motion capture data regarding 3-year-old, 4-year-old, and 5-year-old children in the nursery schools (n = 84) and kindergartens (n = 94) through a three-way non-repeated ANOVA. As a result, a statistically significant difference was observed in movement of body parts. Specifically, right hand movement such as moving distance and the moving average acceleration showed a significance of difference. Secondly, machine learning (decision trees, Sequential Minimum Optimization algorithm (SMO), Support Vector Machine (SVM) and neural network (multilayer perceptron)) was deployed to build classification models for evaluation of degree of musical development classified by educators with simultaneously recorded children's video with associated motion capture data. Among varieties of trained classification models, multilayer perceptron obtained best results of confusion matrix and showed fair classifying precision and usability to support educators to evaluate children's achievement degree of musical development. As a result of the machine learning of multilayered perceptron, the movement of the pelvis has a strong relationship with musical development degree. Its classification accuracy found consistent to affirm the availability to utilize the model to support educators to evaluate children's attainment of musical expression.