Extreme Learning Machine for Active RFID Location Classification

This paper is a preliminary work which seeks the possibilities of using Extreme Learning Machine (ELM) for location classification. We gathered signal strength data from Radio Frequency Identification (RFID) tags and fed the data into the ELM to find in which room a tag is located. We also investigated ELM configuration that results best accuracy for solving our classification problem in terms of the number of training data, regularization factor, the number of time samples, and the number of hidden neurons. Given the problem is to identify in which room a tag is located among 6 rooms by using 2 readers, we achieved 87 percent accuracy with 1 sample, regularization factor C = 2(30), 5 percent training data, and 100 hidden neurons configuration. In simulation-based testing, we found that ELM classification performance is better than LANDMARC performance and comparable with WPL and ELM regression performance with nearest-room coordinate conversion.