ABSTRACT
CONTENTS 1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.1 Computational Topics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 1.2 The Raw Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.1 Processing the Raw Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 1.3 Cleaning the Data and Building a Representation for Analysis . . . . . . . . . . . . . . . 12
1.3.1 Exploring Orientation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 1.3.2 Exploring MAC Addresses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 1.3.3 Exploring the Position of the Hand-Held Device . . . . . . . . . . . . . . . . . . . . . 18 1.3.4 Creating a Function to Prepare the Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.4 Signal Strength Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 1.4.1 Distribution of Signal Strength . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 1.4.2 The Relationship between Signal and Distance . . . . . . . . . . . . . . . . . . . . . . . 26
1.5 Nearest Neighbor Methods to Predict Location . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 1.5.1 Preparing the Test Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 1.5.2 Choice of Orientation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 1.5.3 Finding the Nearest Neighbors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 1.5.4 Cross-Validation and Choice of k . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
1.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
1.1 Introduction The growth of wireless networking has generated commercial and research interests in statistical methods to reliably track people and things inside stores, hospitals, warehouses, and factories. Global positioning systems (GPS) do not work reliably inside buildings, but with the proliferation of wireless local area networks (LANs), indoor positioning systems (IPS) can utilize WiFi signals detected from network access points to answer questions such as: where is a piece of equipment in a hospital? where am I? and who are my neighbors? Ideally, with minimal training, calibration, and equipment, these questions can be answered well in near real-time.
