ランダムフォレスト回帰を用いた認知症徘徊予測モデルの検証

抄録

This report brings a function of sending alert at wandering as an idea. The function uses record of latitude and longitude from GPS terminal with possible person of wandering. To realize the function, daily life movement of non-wandering are learned by machine learning for past location data from GPS terminal. Wandering or non-wandering is judged by GPS terminal real location data and machine learning. For the judgement, the machine learning is used random forest regression to aim to apply theory and develop system simply. The past daily life's movement locations of non-wandering are learned for possible persons of wandering, a movement location of 10 minutes later is predicted by developed program. This study compared a predicted movement location with a real location, when its gap is big, judge as wandering. Explanatory variables in the learning data set were considered by more than one case, for example, suitable number of nearest time points as input. In addition, Correctness of the wanderings detection by this function was considered.