Machine Learning in Wearable Device Fraud Detection
With wearable devices playing a pivotal role in modern healthcare and clinical trials, ensuring the accuracy and authenticity of the data they produce is paramount. This research delves into the application of Machine Learning, especially Deep Learning Image Analysis, to detect potential fraudulent behaviors associated with data generation from wrist-worn wearable devices.
Main Highlights:
Data Authenticity Concerns: There is a growing need for healthcare providers and regulatory authorities to confirm that wearable device data truly originates from the intended patient. Despite this, many current systems fail to reliably link patients with their data, leading to potential fraudulent activities like outsourcing data generation to others or machines.
Previous Findings: Earlier studies have used Machine Learning to analyze raw accelerometer data to detect anomalies, indicating potential fraud. However, this approach mainly focuses on active states such as walking or running.
Novelty of the Study: Departing from previous methodologies, this research centers on heart rate monitor data, specifically photoplethysmography (PPG) data, offering insights even when patients are not particularly active.
Objective: The primary aim is to evaluate whether Deep Learning Image Analysis can identify potential fraudulent behavior during the generation of raw PPG data in clinical or healthcare settings using wrist-worn devices.
Methods: PPG data sourced from PhysioNet was processed and converted into graphical representations. The Inception-v3 Machine Learning model, originally designed for image classification tasks, was adapted to this challenge, classifying data epochs to individual persons.
Encouraging Results: The results are promising:
Over 80% accuracy in training was achieved in a few cycles.
On evaluation, the model correctly predicted the origin of 87.3% of the data plots.
A confusion matrix reflected prediction accuracies ranging from 75% to 97% across the six datasets.
Conclusively, Machine Learning, especially Deep Learning Image Analysis, emerges as a potent tool in authenticating data from wearable devices, instilling more confidence in their use in critical sectors like healthcare.
