AI-Driven Error Correction for KNN-Based RFID Localization

Radio Frequency Identification (RFID) localization has been studied constantly for the last decade. New machine learning and artificial intelligence methods are constantly being developed to reduce error and make implementation of different applications more likely. This work focuses on reducing error from preliminary results obtained in a previous work by using basic Artificial Neural Networks (ANNs). Results from a moving 3-Dimensional kNN algorithm are processed in different stages and different features are fed into the ANN from the previous work. The goal is to train the network to try and predict the error obtained from the kNN algorithm. In general, the predictions that stem off the ANN can be fed back into the predictions coming from the k NN algorithm, thus reducing the error. We find that simple ANNs can achieve a Mean Absolute Error (MAE) as low as 3.4cm, depending on antenna reception level thresholds, compared to 10.5cm compared in our previous work.