For a large sensor network with severe energy constraints, it is infeasible for all sensor nodes to continuously send all sensor readings to the base station to answer continuous monitoring queries. In order to reduce the energy consumption caused by data transmission and thus prolong the network lifetime, various approximate query processing techniques have been proposed. One of the well-known approaches for approximate query processing is to install a filter at each sensor node so that a reading of a sensor node is prevented from being sent to the base station if the value of the reading is within the range specified by the filter. However, most existing methods set and maintain the filters under the assumption that the next reading of a sensor node will be the same as the previous one. In this paper, we propose a new filter setting method that dynamically adjusts the filters when the readings of sensor nodes change in a periodically repeating pattern. The proposed method predicts the next readings of sensor nodes based on historical sensor data and adaptively adjusts the filters using the predicted values. Simulation-based experimental results show that the proposed method significantly reduces the number of message transmissions compared with the previous ones.