<p>For the people who needs continuous care and nursing, unconstraint vital sign monitoring will take important role in their quality of life. The system should classify the users posture exactly and grasp precise position of body parts to monitor the patient's vital sign without attaching sensors to body. In this study, in-bed posture analytics are done for recognition of patient's 5 kind of pose including fowler's position from pressure sensor mattress data. In-bed posture recognition was carried out by convolutional neural network models with 5,907 data that obtained from pressure sensor mattress. This study mainly focused on gradient descent algorithm and comparing to choose its best optimizer to prevent bad performance resulted by poor saddle point of parameters. By optimizing the model, the deep learning model reached 100% classification accuracy.</p>