Objective comparison of audiometric profile frameworks across large-scale datasets.

Area of Science:

• Audiology and Hearing Science
• Data Science and Machine Learning

Background:

• Audiometric profiles categorize hearing loss patterns from audiograms.
• Existing profiling frameworks lack systematic evaluation of dataset influence on performance.

Purpose of the Study:

• To compare the structural performance of six audiometric profiling frameworks.
• To assess the impact of dataset characteristics on framework generalizability.

Main Methods:

• Six established audiometric profiling frameworks were analyzed.
• Five large-scale datasets from the US and Germany were utilized.
• Davies-Bouldin score and principal component analysis were employed for evaluation.

Main Results:

• Clustering performance, measured by the Davies-Bouldin score, was broadly comparable across datasets.
• Specific audiometric profiles exhibited dataset-dependent variations in performance.
• Framework robustness and generalizability were assessed across diverse samples.

Conclusions:

• Audiogram-based classification frameworks demonstrate generalizable performance across large datasets.
• Dataset characteristics can influence the performance for specific hearing loss profiles.
• Findings aid in selecting robust classification methods for diverse populations.