---
title: "Why ML-Based Cough Models Do Not Generalize: A Systematic Cross-Dataset Evaluation for Tuberculosis Screening"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2608.25846v1"
permalink: "https://feed.astralor.com/items/itm_ed3501865e3baff64a302bf5"
published: "2026-08-26T14:22:32.000Z"
score: 73
channel: "finance"
tags: ["arXiv","CODA","Cough Acoustics","Machine Learning","Deep Learning","TB Screening"]
---

# Why ML-Based Cough Models Do Not Generalize: A Systematic Cross-Dataset Evaluation for Tuberculosis Screening

## AI commentary

> This paper is a must-read for anyone working in medical AI audio analysis—it shatters the illusion that high internal accuracy is enough with a simple three-dataset validation. For teams building cough screening products, device diversity and external validation are not optional; they are a matter of survival.

## At a glance

This paper systematically evaluates the cross-dataset generalizability of machine learning and deep learning cough-based tuberculosis screening classifiers across three independent datasets. Despite moderate within-dataset performance (ROC-AUC up to 0.755 ± 0.056), external performance frequently drops below 0.6, indicating models largely learn artifacts of recording devices and data collection rather than disease-related acoustic features. Audio representations cluster by device and dataset rather than TB status; device mismatch degrades transfer, while device-diverse training improves generalization. A clinical-variable baseline generalizes more consistently (ROC-AUC 0.655-0.711), suggesting acquisition-specific variability is a stronger driver of poor generalizability than population shift. The study emphasizes that external validation is essential before cough-based TB models are clinically ready.

## Key points

- A systematic evaluation across three independent datasets shows that cough-based TB classifiers generalize poorly, with external ROC-AUC frequently below 0.6.
- Audio representations cluster by recording device and dataset rather than TB status, indicating models capture acquisition artifacts rather than disease signals.
- Device mismatch significantly degrades transfer, while device-diverse training improves generalization; a clinical-variable baseline outperforms acoustic models in stability.
- The study warns that high within-dataset performance is insufficient for clinical deployment; external validation is a necessary prerequisite.

## Full text

Cough acoustics are promising for non-invasive tuberculosis (TB) screening, yet whether machine learning (ML) models capture disease-related acoustics or artifacts of data collection remains unresolved. We evaluated the cross-dataset generalizability of classical ML and deep learning (DL) cough-based TB classifiers across three independent datasets. Despite moderate within-dataset performance (ROC-AUC up to $0.755 \pm 0.056$), both pipelines fail to generalize, with external performance frequently below 0.6, indicating a possible limitation of the data. We further observed audio representations are organized by recording device and dataset rather than TB status, predicted TB probability tracks country-level prevalence in CODA, and device mismatch degrades transfer while device-diverse training improves it. Additionally, a clinical-variable baseline generalizes more consistently (ROC-AUC $0.655 - 0.711$), indicating acquisition-specific variability is a stronger driver of poor generalizability than population shift. High within-dataset performance is not enough. External validation is essential before cough-based TB models are clinically ready.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_ed3501865e3baff64a302bf5