A comparison of decision tree based techniques for indoor positioning system

Currently, an indoor positioning system based on a fingerprint technique for wireless networks under IEEE 802.11 standard uses a method to collect a received signal strengths (RSS) to create a radio map in an offline phase. Then, it detects the RSS in an online phase to compare and find the position. However, while detecting and gathering the RSS, there are some variations of the RSS that affect the accuracy of position estimation. Therefore, there are several methods used to estimate the position in order to improve the accuracy, but the one focused in this paper is a decision tree based classification. The decision tree based classification method is found that it can provide better improvement in accuracy than the others, e.g., K-Nearest Neighbor (K-NN), Bayesian, and Neural networks. However, the techniques used to construct the decision tree are varied depending on the algorithm used to implement. Therefore, this paper is a comparison of decision tree based techniques using typical decision tree (DT) and Gradient boosted tree algorithms for estimating the position indoor. In the study, the RSSs collected from access points in the experimental area are used as the training and testing data. The decision tree models are created by using typical DT and Gradient boosted algorithms based on the training data obtained. There are two factors to consider in the comparative study, i.e., the number of training data and the number of reference radio signals. The testing results from the experiment showed that the decision tree based on Gradient boosted algorithm yielded more accurate results than typical DT, where the amount of 19 reference radio signals and 50 samples of training data gave the best result.