Abstract:
Abstract: Indoor positioning systems are becoming more popular due to the limitations of Global Positioning System (GPS) in locating
a person’s location in a closed environment. One of the most frequently used method for indoor positioning systems is fingerprinting.
Fingerprinting relies on obtaining a set of Received Signal Strength Indicator (RSSI) values from multiple access points and then
comparing it to the database to predict a device location using methods such as k-Nearest Neighbors (KNN). However, this process
takes a long time and does not scale well with large and high dimensional dataset. In this study, we evaluate and compare three different
methods for search optimization which are Hierarchical Navigable Small World (HNSW), Locality-Sensitive Hashing (LSH) and
Inverted File Index (IVF). The evaluation will be conducted based on the search speed and the number of correct predictions of each
searching method. Our results show that HNSW outperforms the other methods by a slight margin in terms of accuracy, giving an
accuracy of 74.1% compared to LSH accuracy of 72.5% and IVF accuracy of 73.2%. In terms of search speed, HNSW and IVF are
significantly faster than LSH, with average time per query of 0.014 and 0.016 seconds respectively, compared to LSH time of 4.813
seconds per query
