[Paper Review] Investigation into respiratory sound classification for an imbalanced data set using hybrid LSTM-KAN architectures

Nithinkumar K., Anand R.|arXiv (Cornell University)|Jan 7, 2026

Phonocardiography and Auscultation Techniques0 citations

TL;DR

The paper proposes a hybrid LSTM-KAN model with focal loss, data augmentation, and SMOTE to classify imbalanced respiratory sounds, achieving 94.6% accuracy and macro F1 of 0.703 on a six-class dataset dominated by COPD.

ABSTRACT

Respiratory sounds captured via auscultation contain critical clues for diagnosing pulmonary conditions. Automated classification of these sounds faces challenges due to subtle acoustic differences and severe class imbalance in clinical datasets. This study investigates respiratory sound classification with a focus on mitigating pronounced class imbalance. We propose a hybrid deep learning model that combines a Long Short-Term Memory (LSTM) network for sequential feature encoding with a Kolmogorov-Arnold Network (KAN) for classification. The model is integrated with a comprehensive feature extraction pipeline and targeted imbalance mitigation strategies. Experiments were conducted on a public respiratory sound database comprising six classes with a highly skewed distribution. Techniques such as focal loss, class-specific data augmentation, and Synthetic Minority Over-sampling Technique (SMOTE) were employed to enhance minority class recognition. The proposed Hybrid LSTM-KAN model achieves an overall accuracy of 94.6 percent and a macro-averaged F1 score of 0.703, despite the dominant COPD class accounting for over 86 percent of the data. Improved detection performance is observed for minority classes compared to baseline approaches, demonstrating the effectiveness of the proposed architecture for imbalanced respiratory sound classification.

Motivation & Objective

• Address the challenge of severe class imbalance in respiratory sound datasets.
• Develop a hybrid architecture combining LSTM for sequential feature encoding with a Kolmogorov-Arnold Network (KAN) for classification.
• Integrate feature extraction with imbalance mitigation techniques to improve minority class recognition.
• Evaluate performance on a public six-class respiratory sound dataset with a COPD-dominated distribution.

Proposed method

• Use LSTM to encode sequential audio features.
• Pair LSTM with a Kolmogorov-Arnold Network (KAN) for classification.
• Apply focal loss to address class imbalance.
• Incorporate class-specific data augmentation strategies.
• Apply SMOTE to balance minority classes.
• Evaluate on a six-class respiratory sound dataset with skewed distribution.

Experimental results

Research questions

• RQ1 Can a hybrid LSTM-KAN architecture improve classification of imbalanced respiratory sound data compared to baselines?
• RQ2 How do imbalance mitigation techniques (focal loss, augmentation, SMOTE) affect minority class performance?
• RQ3 What is the overall accuracy and macro-averaged F1 on a six-class, COPD-dominated respiratory sound dataset?

Key findings

• Hybrid LSTM-KAN achieves 94.6% overall accuracy.
• Macro-averaged F1 score reaches 0.703.
• Minority class detection improves relative to baseline approaches.
• The COPD class constitutes over 86% of the data, highlighting severe imbalance.
• Imbalance mitigation techniques contribute to better recognition of minority classes.