Machine Learning-Based Area Estimation Using Data Measured Under Walking Conditions

Abstract

This study examines the accuracy and measurement costs associated with room-level indoor-area estimation using a wireless LAN. Utilizing fingerprinting, a method that compares user-measured access point (AP) information with pre-existing AP data from service providers, this study introduces a cost-effective approach. Our proposed machine learning (ML)-based method leverages data collected by users while traversing different locations within an area, thereby significantly reducing the measurement time. Furthermore, this study contrasts the effectiveness of convolutional neural networks (CNN) and support vector machines (SVM) in area estimation using this novel measurement technique. Both CNN and SVM demonstrated comparable accuracy, with SVM exhibiting a shorter processing time.