Wi-Fi Router Signal Coverage Position Prediction System using Machine Learning Algorithms

Abstract

As a popular topic, indoor positioning has gradually drawn the attention of both academia and business. Numerous location based services including healthcare, repository tracking, and security call for accurate estimation. An accurate estimation might be achieved by using additional location-sensing equipment, but this is not generally done because it would result in expensive brand specialization. A flexible and affordable location determination technique that exploits the already-existing WLAN infrastructure in indoor spaces has been designed without incurring additional costs, among all suggestions in the literature that include hardware and highly complex computations.This positioning strategy is becoming more popular. Soon in actual surroundings, WLAN will be able to be employed as part of an indoor positioning system. In comparison to similar systems, it is a good option in terms of accuracy, precision, and cost. It has also become the most user-friendly way, particularly with the widespread use of smartphones and tablet computers. In the literature, many machine learning algorithms such as cluster-filtered KNN and fuzzy c means algorithms have been proposed for the WiFi router signal coverage position prediction system but the disadvantage of these approaches is that it will consume much pre-processing time for reference points data and also these approaches found to be less accurate. In the proposed study, adaptive KNN-based machine learning model has been proposed for the WiFi signal coverage prediction system. The proposed method adjusts the value of K for each position by examining the relationship between the K value and the intensity of the received WiFi signal. This technique improves positioning accuracy by more than 30% when compared to the existing approach. The experimental results are conducted by applying various machine learning algorithms. The experimental finding demonstrates that the proposed approach obtained better results compared to traditional algorithms.