The accuracy of facial recognition technology using AI is improving, and a lot of research is progressing on estimating emotions from facial expressions. However, emotion estimation for elderly people has shown that judging emotions is more difficult than for the ordinary people due to poor facial expressions caused by weakened facial muscles. Therefore, in this paper, we use the Japanese elderly database to examine a feature point determination method suitable for reading six emotions (joy, sadness, surprise, fear, anger, dis- gust) from elderly face images.“difficulty in extracting facial features from elderly individuals with diminished facial muscle activity”, we conducted facial feature point selection using MediaPipe’s Faceesh and Genetic Algorithm. Experimental results show GA-based facial feature point selection make classification accuaracy improvement and some issues to be tackled in further improvement are identified.