Cough Classification For Early Detection Of Respiratory Diseases Using A Combination Of Mel-Frequency Cepstral Coefficients (MFCC) And Capsule Network

Andhika, Desta Rizky (2026) Cough Classification For Early Detection Of Respiratory Diseases Using A Combination Of Mel-Frequency Cepstral Coefficients (MFCC) And Capsule Network. Undergraduate thesis, Universitas Pembangunan Nasional Veteran Jawa Timur.

	Text (cover) cover_skripsi_new.pdf Download (1MB)
	Text (bab 1) bab1_skripsi.pdf Download (461kB)
	Text (bab 2) bab2_skripsi.pdf Restricted to Repository staff only until 3 September 2028. Download (1MB)
	Text (bab 3) bab3_skripsi.pdf Restricted to Repository staff only until 3 September 2028. Download (1MB)
	Text (bab 4) bab4_skripsi.pdf Restricted to Repository staff only until 3 September 2028. Download (5MB)
	Text (bab 5) bab5_skripsi.pdf Download (489kB)
	Text (daftar pustaka) daftarpustaka_skripsi.pdf Download (466kB)

Abstract

Respiratory diseases such as acute respiratory infections (ARI), pneumonia, and chronic obstructive pulmonary disease (COPD) remain major global causes of morbidity, with cough serving as a crucial early clinical indicator. In conventional medical practice, distinguishing between dry (non-productive) and wet (productive) coughs is often subjective. This study aims to develop an artificial intelligence-based cough classification system as a non-invasive early screening tool for respiratory conditions using Mel-Frequency Cepstral Coefficients (MFCC) feature extraction combined with a Capsule Network (CapsNet) architecture. The methodology began by curating a public Kaggle dataset consisting of 27,550 raw metadata entries into a balanced, high-confidence experimental subset of 800 verified audio samples (400 dry and 400 wet). The preprocessing pipeline involved WAV conversion, mono reduction, 16 kHz resampling, silence trimming, adaptive RMS energy-based cough event segmentation, and duration standardization to 3 seconds (48,000 samples). Spectral features were extracted as MFCC coefficients and normalized using Z-scores. Model evaluation encompassed ten experimental scenarios (S01–S10), each run repeatedly across 10 runs with distinct random seeds (seeds 42–51) to assess model stability and performance. Experimental results demonstrated that scenario S02 (MFCC13-only, 2 routing iterations, class capsule dimension of 16, dropout rate of 0.40, and primary capsule stride of 3x8) achieved the highest and most stable average performance among all single and ensemble scenarios. Scenario S02 achieved a mean Balanced Accuracy of 71.67% ± 2.08%, an Accuracy of 71.67%, an F1-Score of 69.77%, a Precision of 71.00%, a Recall of 65.15%, a Specificity of 77.83%, a Macro-F1 of 71.45%, and a Matthews Correlation Coefficient (MCC) of 0.4384. The optimal model was subsequently deployed into a web-based screening prototype built with Streamlit. This research confirms that combining MFCC with Capsule Networks effectively captures the hierarchical spectral relationships of cough sounds to support AI-driven medical prescreening. Keywords: Cough Classification,Early Detection of Respiratory Diseases, MFCC, Capsule Network, Dynamic Routing

Item Type:

Thesis (Undergraduate)

Contributors:

Contribution	Contributors	NIDN/NIDK	Email
Thesis advisor	Muttaqin, Faisal	NIDN0030058602	faisalmuttaqin.if@upnjatim.ac.id
Thesis advisor	Anggraeny, Fetty Tri	NIDN0711028201	fettyanggraeny.if@upnjatim.ac.id

Subjects:

Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Q Science > QA Mathematics > QA76.6 Computer Programming

Divisions:

Faculty of Computer Science > Departemen of Informatics

Depositing User:

Desta Rizky Andhika

Date Deposited:

07 Sep 2026 02:09

Last Modified:

07 Sep 2026 02:45

URI:

https://repository.upnjatim.ac.id/id/eprint/59960