Statistical Analysis of Daily Life Data Acquired by a Sensor-embedded Plushie Device for Understanding the Behavior of the Terrible Twos

The burden of childcare for the terrible twos is enormous and has become a serious problem. Therefore, a balance between reducing the burden of childcare and providing detailed supervision of toddlers is required. A plushie device was developed that can safely measure the toddler's behaviors for a long time, including externalizing behaviors such as throwing and hitting. This study aims to elucidate the characteristics of sensor responses from data that reflect the toddler's state and behavior, and to demonstrate an analysis method that makes it possible to understand the relationship with the toddler's behavior. Using the developed plushie, we experimented to measure the behavior of the terrible twos in the home. The acquired data is visualized by creating a histogram for each sensor and compared with the results from the guardian questionnaire. We suggest the acceleration sensor reacts to any movement, such as car travel, whereas the bending sensor reacts to the intentional contact of a toddler, such as playing. Both the frequency and intensity of contact decreases when the toddler is unhealthy or in a bad mood. Consequently, car travel actions are successfully labeled using machine learning while showing the usefulness of appropriately selecting sensors and statistics for analysis.