KLASIFIKASI SUARA JANTUNG MENGGUNAKAN JARINGAN NEURAL DENGAN CIRI STATISTIS DAN SPEKTRAL
| Main Authors: | , Domy Kristomo, , Dr. Indah Soesanti, ST., MT. |
|---|---|
| Format: | Thesis NonPeerReviewed |
| Terbitan: |
[Yogyakarta] : Universitas Gadjah Mada
, 2014
|
| Subjects: | |
| Online Access: |
https://repository.ugm.ac.id/133511/ http://etd.ugm.ac.id/index.php?mod=penelitian_detail&sub=PenelitianDetail&act=view&typ=html&buku_id=74198 |
Daftar Isi:
- Feature extraction becomes a very important part in exploring the dynamics of the data contained in the heart sound. In particular to distinguish the normal heart sounds and different types of murmurs. Therefore it is necessary to learn effective feature extraction method. To improve the prediction of accuracy and minimize the computation time, feature selection is proposed. Artificial Neural Network (ANN) is a crucial tool in identification and classification. The aim of this research is to apply ANN for classify normal heart sound and murmurs. In this research has developed a heart sound feature extraction method using statistical approach, namely by calculating the mean value, standard deviation, entropy, skewness, and kurtosis of heart sound in time and frequency domain. The advantage statistical method is the feature extraction can be done in time domain, with the result that computation is faster, because there isn�t transformation. In this research, the correlation based feature selection (CFS) is proposed to select the best feature as the input for classifier. Neural network backpropagation using training algorithm Levenberg Mardquart is proposed for learning and testing of statistical features. The result using Neural Network shows that statistical feature in time domain gives better classification performance than statistical feature in frequency domain. The best classification performance is resulted from composite and feature selection of statistical time domain feature, frequency domain feature, RMS, Shannon Energy Feature. Statistical method using composite of time and frequency feature consisting 10 feature gives better performance than PSD Welch method which using 17 and 30 sampling. Keywords : Feature extraction, heart sound, statistical, mean, standard deviation, entropy, skewness, kurtosis, welch, feature selection