Studying Accuracy of Machine Learning Models Trained on Lab Lifting Data in Solving Real-World Problems Using Wearable Sensors for Workplace Safety

TL;DR

This paper investigates the challenges of applying lab-trained machine learning models to real-world workplace safety scenarios using wearable sensors, identifying causes of performance drops and proposing solutions to improve accuracy.

Contribution

It introduces four novel methods to enhance the transferability of lab-trained models to real-world environments in workplace safety applications.

Findings

Performance drops significantly when deploying lab-trained models in real-world settings.

Identified key factors causing model failure in real-world data.

Proposed solutions improve model accuracy in practical applications.

Abstract

Porting ML models trained on lab data to real-world situations has long been a challenge. This paper discusses porting a lab-trained lifting identification model to the real-world. With performance much lower than on training data, we explored causes of the failure and proposed four potential solutions to increase model performance