Enhancing Machine Learning Performance with Continuous In-Session Ground Truth Scores - Pilot Study on Objective Skeletal Muscle Pain Intensity Prediction

Aug 5, 2026

Upload by Anda

Machine learning models trained on subjective self-report pain scores face difficulty producing objective, accurate pain classification because recorded values often diverge from real-time pain experiences. This pilot study builds devices for continuous in-session pain scoring and simultaneous acquisition of ANS-modulated endodermal activity (EDA). Data from 24 subjects undergoing post-exercise circulatory occlusion with stretch supported extraction of time-domain EDA features and in-session ground truth, with post-experiment VAS collected for comparison.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages18

Reading Time45 min

AccessFree to download

Introduction

Background: opioid crisis and pain measurement limitations

Need for objective pain assessment

Motivation: ground truth sparsity and label noise

Study Overview

Continuous in-session pain scoring devices

Data collection protocol (N=24, PECO with stretch)

Feature extraction from time-domain EDA

Model Training and Evaluation

Multi-layer Perceptron (MLP) vs Random Forest (RF)

10-fold cross-validation and performance comparison

Findings and Implications

Continuous in-session ground truth improves ML performance

Reducing ground truth sparsity and variance

Conclusion

Guidance for future objective-based ML pain systems

Enhancing Machine Learning Performance with Continuous In-Session Ground Truth Scores: Pilot Study on Objective Skeletal Muscle Pain Intensity Prediction Boluwatife E. Faremi 1,*, Jonathon Stavres 2, Nuno Oliveira 2, Zhaoxian Zhou 1 and Andrew H. Sung 1 1 School of Computing Sciences and Computer Engineering, University of Southern Mississippi, Hattiesburg, MS 39401, USA 2 School ofKinesiology and Nutrition, University of Southern Mississippi, Hattiesburg, MS 39401, USA * [Correspondence: boluwatife.faremi@usm.edu](Correspondence: boluwatife.faremi@usm.edu) Abstract: Machine learning (ML) models trained on subjective self-report scores struggle to objectively classify pain accurately due to the significant variance between real-time pain experiences and recorded scores afterwards. This study developed two devices for acquisition of real-time, continuous in-session pain scores and gathering ofANSmodulated endodermal activity (EDA).The experiment recruited N = 24 subjects who underwent a post-exercise circulatory occlusion (PECO) with stretch, inducing discomfort. Subject data were stored in a custom pain platform, facilitating extraction of time-domain EDA features and in-session ground truth scores. Moreover, post-experiment visual analog scale (VAS) scores were collected from each subject. Machine learning models, namely Multi-layer Perceptron (MLP) and Random Forest (RF), were trained using corresponding objective EDA features combined within-session scores and post-session scores, respectively. Over a 10-fold cross-validation, the macro-averaged geometric mean score revealed MLP and RF models trained with objective EDA features and in-session scores achieved superior performance (75.9% and 78.3%) compared to models trained with post-session scores (70.3% and 74.6%) respectively. This pioneering study demonstrates that using continuous in-session ground truth scores significantly enhances ML performance in pain intensity characterization, overcoming ground truth sparsity-related issues, data imbalance, and high variance. This study informs future objective-based ML pain system training. Keywords: Opioids, Pain, Machine learning, EDA, Objective, Ground truth, Self-report, VAS, Autonomic Nervous system (ANS) 1. Introduction Pain is reported to be the center cog of opioid crisis and the majority of those suffering from opioid use disorder (OUD) state it all started from using opioids for pain with legitimate prescription [1] . In [2], authors reveal that about 9.7 million individuals aged 12 or above have misused prescribed opioids, leading to the center for disease control (CDC) declaring an opioid epidemic. Nonetheless, opioid remains the frontline treatment in the United States irrespective of several clinical guidelines recommending non-pharmacological alternatives [2, 3, 4] . In [2], authors divulge that orthopedic pain (34.8 %) is a primary reason for opioid dispensing trailed by dental pain (17.3 %), back pain (14.0 %) and headache (12.9 %) . When abused, harms arising from opioid intake are but not limited to overdose, addiction, diversion, and death [1, 5] . These side effects make pain measurement clinically relevant by medical practitioners to prevent the absolute risk of death [6, 7] . Consequently, pain and its intensity are described as a personal experience that can only be described by the individual experiencing pain and its often assessed by pain intensity rating scales [8]. Typically, traditional pain rating or self-report scales such as visual analog scales (VAS), numerical rating scales (NRS) are acknowledged to be highly subjective, prone to anchor drift over time and significant variation in reported scores [6, 7, 8] . Additionally, these scales are notorious for misleading practitioners into over and under prescribing opioids, encouraging catastrophizing amongst patients, and sheltering of social-ecological construct such as race, implicit and explicit bias amongst dishonest healthcare

FAQs

Why do models trained on self-report pain scores struggle with objective pain classification?

Self-report scores differ from real-time pain experiences, introducing high variance and label noise that reduces predictive reliability.

What does the pilot study collect during the experiment?

It collects continuous in-session pain ground truth scores and autonomic nervous system activity signals via EDA during post-exercise circulatory occlusion with stretch, along with post-experiment VAS.

How do MLP and RF models trained with different ground truth sources compare?

Across 10-fold cross-validation, models using objective EDA features with in-session scores achieved higher macro-averaged geometric mean performance (75.9% MLP, 78.3% RF) than models trained with post-session scores (70.3% MLP, 74.6% RF).

Aug 5, 2026

Upload by Anda