Q-Learning-Based Pending Zone Adjustment for Proximity Classification

This paper presents a Q-learning-based pending zone adjustment for received signal strength indicator (RSSI)-based proximity classification (QPZA). QPZA aims to improve the accuracy of RSSI-based proximity classification by adaptively adjusting the size of the pending zone, taking into account changes in the surrounding environment. The pending zone refers to an area in which the previous result of proximity classification is maintained and is expressed as a near boundary and a far boundary. QPZA uses Q-learning to expand the size of the pending zone when the noise level increases and reduce it otherwise. Specifically, it calculates the noise level using the estimation error of a device deployed at a specific location. Then, QPZA adjusts the near boundary and far boundary separately by inputting the noise level into the near and far boundary adjusters, consisting of the Q-learning agent and reward calculator. The Q-learning agent determines the next boundary using the Q-table, and the reward calculator calculates the reward using the noise level. QPZA updates the Q-table of the Q-learning agent using the reward. To evaluate the performance of QPZA, we conducted an experimental implementation and compared the accuracy of QPZA with that of the existing approach. The results showed that QPZA achieves 11.69% higher accuracy compared to the existing approach, on average.