[Paper Review] LOCATER: Cleaning WiFi Connectivity Datasets for Semantic Localization

Yiming Lin, Daokun Jiang+5 more|arXiv (Cornell University)|Apr 20, 2020

Indoor and Outdoor Localization Technologies62 references2 citations

TL;DR

LOCATER is a data cleaning system that transforms raw WiFi connectivity logs into accurate semantic indoor locations (e.g., rooms, floors) by treating coarse localization as a missing value detection problem and fine-grained localization as a probabilistic disambiguation task. It achieves high accuracy using semi-supervised learning and probabilistic inference on historical WiFi data without additional hardware or user cooperation.

ABSTRACT

This paper explores the data cleaning challenges that arise in using WiFi connectivity data to locate users to semantic indoor locations such as buildings, regions, rooms. WiFi connectivity data consists of sporadic connections between devices and nearby WiFi access points (APs), each of which may cover a relatively large area within a building. Our system, entitled semantic LOCATion cleanER (LOCATER), postulates semantic localization as a series of data cleaning tasks - first, it treats the problem of determining the AP to which a device is connected between any two of its connection events as a missing value detection and repair problem. It then associates the device with the semantic subregion (e.g., a conference room in the region) by postulating it as a location disambiguation problem. LOCATER uses a bootstrapping semi-supervised learning method for coarse localization and a probabilistic method to achieve finer localization. The paper shows that LOCATER can achieve significantly high accuracy at both the coarse and fine levels.

Motivation & Objective

• To address the challenge of transforming noisy, incomplete WiFi connectivity logs into meaningful semantic indoor locations such as rooms or floors.
• To frame semantic localization as a data cleaning problem rather than a direct positioning task.
• To eliminate reliance on specialized hardware or user cooperation by leveraging existing WiFi infrastructure and historical connectivity patterns.
• To improve accuracy in coarse and fine-grained localization using bootstrapping semi-supervised learning and probabilistic inference.
• To support smart space applications like occupancy monitoring, thermal control, and crowd density tracking in buildings.

Proposed method

• Treats the assignment of a device to a WiFi access point (AP) between connection events as a missing value detection and repair problem.
• Uses a bootstrapping semi-supervised learning method to infer coarse semantic locations (e.g., floor or region) based on temporal patterns in connectivity events.
• Applies a probabilistic model to resolve fine-grained location disambiguation (e.g., which room) using spatial and temporal correlations across devices.
• Leverages historical WiFi event logs to estimate confidence intervals for inter-connection durations, identifying in-building vs. out-of-building states.
• Employs a possible worlds model to compute posterior probabilities of room assignments based on observed device connectivity patterns.
• Optimizes for scalability and near real-time performance through efficient data processing and parameter tuning using cross-validation or empirical distribution fitting.

Experimental results

Research questions

• RQ1 Can WiFi connectivity data be cleaned to enable accurate semantic localization without additional hardware or user cooperation?
• RQ2 How can missing AP associations between connectivity events be effectively detected and repaired?
• RQ3 To what extent can semi-supervised learning improve coarse localization accuracy using only connectivity logs?
• RQ4 How can probabilistic inference enhance fine-grained localization (e.g., room-level) from sparse and noisy WiFi data?
• RQ5 Can the system scale to real-world datasets while maintaining high accuracy and low latency?

Key findings

• LOCATER achieves significantly high accuracy in both coarse and fine-grained semantic localization using only raw WiFi connectivity logs.
• The system outperforms baseline heuristics such as spatial overlap or random selection in room-level localization.
• Parameter tuning via confidence intervals on average connection durations improves detection of in-building vs. out-of-building states with 95% confidence.
• The bootstrapping semi-supervised learning approach effectively learns coarse location patterns from limited labeled data.
• The probabilistic disambiguation model successfully resolves ambiguous room assignments by leveraging device co-occurrence and spatial constraints.
• LOCATER demonstrates scalability and near real-time performance, making it suitable for production deployment in smart buildings.