In this study, a bed-leaving prediction system with light weight and thin shaped load sensor was prototyped. The prototype system could obtain the load within the error of 10% and estimate center-gravity of a bed within the error of 38mm. Then, we attempted to detect the posture change from moving behavior of the center-gravity of the bed. In general, a change of posture causes a large change of center-gravity, the detection of posture change should eliminate the small fluctuation of center-gravity caused by small movement of body. The fluctuation of center-gravity could be separated from posture change by using a discriminant with threshold of position change and time change of the center-gravity. As result, the detection error was effectively reduced. In addition, the bed-leaving prediction system was improved to wireless communication between the load sensor and a monitor unit for easy-to-use. The load data of the bed could be transmitted to the monitor unit.