In the rehabilitation for the motor function recovery, the menu of the rehabilitation has been decided based on doctor's subjectivity by knowledge and the experience. However, the decision based on doctor's subjectivity has the possibility of inviting the result that the judgement is different in each doctor. On the other hand, authors were relating “muscular power coefficient”, that showed the relation of EMG (Electromyogram) and muscular load, and bodily features in the nonlinear model by GP (Genetic Programing). As a result, a muscular condition was enabled to be estimated objectively only by knowing patient's bodily features. This paper proposed the improvement of the estimation accuracy by clustering based on subject's bodily features, and constructing the nonlinear model in each class. The effectiveness of the proposed method is shown by the experimental work using the subjects at two or more ages.