Burglar Detection System Using Wi-Fi Based Secure Indoor Positioning System

Indoor positioning systems have seen significant development in recent years. Unlike the widely known Global Positioning System (GPS), this research focuses on indoor positioning, commonly referred to as the Indoor Positioning System (IPS). IPS is particularly influential in the digital era, given the advancements in infrastructure and digitalization sectors. These developments often introduce new challenges, such as difficulties in locating people or objects within skyscrapers due to various obstacles. Additionally, threats can emerge from both system vulnerabilities and physical intrusions. This study aims to compare different models based on an existing dataset and a newly collected dataset from a controlled indoor environment. The comparison employs several machine learning algorithms to determine which algorithm achieves the highest accuracy. Furthermore, an attack test method is designed to develop a robust model. From this comparison, the model with the best accuracy will be selected to create a detection system prototype. This prototype will send warning messages via email upon detecting anomalies, potentially proving useful in restricted spaces. The research utilizes Channel State Information (CSI) datasets. Initial testing with various models achieved an accuracy of 74.02% using the random forest algorithm. However, implementing a Decision Tree Attack reduced the accuracy to 52%. Applying a maintenance method using feature squeezing improved model durability, resulting in an accuracy of 68%. The Anomaly Detector implementation successfully sent danger messages via email, demonstrating the practical application of the system.