Stick exercises, which have been attracting attention for improving the health of the elderly, are usually performed in nursing homes under the guidance of nursing staff. However, in the current pandemic in which the elderly are advised to refrain from going out unnecessarily, it is desirable for each individual to be able to perform the stick exercises alone. In this study, we aim to develop a stick exercise support system that can automatically record the number of times an elderly person performs each type of stick exercise and provide feedback to improve the movement for each exercise. As a first step toward the realization of this stick exercise support system, we investigated a method for recognizing exercise movements using inertial measurement unit (IMU) sensors. In the evaluation experiment, 21 subjects performed 3 sets (10 times per set) of eight basic stick exercises. The exercise movements were classified based on the linear acceleration and quaternion data obtained from the IMU. As a result, 90% of F-measure was achieved when using Light GBM as the learning algorithm.