The purpose of study is to propose a task selection algorithm that both keeps information quality and saves power consumption in IoT energy harvesting devices. The proposed algorithm not only keeps stable information quality but saves power loss also. The sensor node operation is divided into four tasks depending on the input data including battery capacity, solar panel charging current, and input sensor data variation. The task selector based on a neural network consists of an input layer, a hidden layer of 20 neurons, and an output layer. The proposed algorithm is different from the predefined task algorithm, which mainly focused on deep sleep mode or scheduled tasks. Our proposed algorithm helps the sensor node to be more adaptive to the environment based on real-time execution at each node. The collected information amount varies according to the input data variation. The experiment results show that the proposed algorithm collects higher quality information at large input data variation. The battery lifetime is also improved by up to 22%.