Adaptive Filter Untuk Deteksi Data Outlier Pada Sinyal Electronic Nose

Purbawa, Doni Putra (2022) Adaptive Filter Untuk Deteksi Data Outlier Pada Sinyal Electronic Nose. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6025201014-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2024. Download (11MB) | Request a copy

Abstract

Electronic Nose (E-nose) merupakan alat peniru sistem kerja hidung yang murah, mudah dan akurat dalam proses implementasi metode machine learning. Beberapa peneliti menggunakan E-nose untuk mendeteksi keaslian daging sapi dicampur dengan babi dan deteksi kualitas makanan. Selain itu, pada bidang medis E-nose juga banyak digunakan untuk penelitian, seperti mendeteksi dini pasien diabetes, kanker paru-paru dan tuberkulosis melalui embusan napas, serta mendeteksi penyakit pernapasan menular (SARS) melalui bau keringat pada ketiak.
Tantangan utama dari penggunaan E-nose yaitu outlier atau data yang berada di luar batas populasi. Outlier merupakan kasus ekstrem yang berbeda dengan observasi lainnya dan keberadaan outlier dapat merusak kualitas data. Pada kasus deteksi penyakit diabetes dan TBC melalui embusan napas, bau makanan atau minuman yang dikonsumsi dapat mengganggu proses deteksi. Sama halnya dengan deteksi penyakit pernapasan menular melalui bau keringat ketiak. Penggunaan deodoran dan parfum serta bau di sekitar ruang sampel dapat merusak kualitas data.
Penelitian ini mengusulkan filter adaptif menggunakan deep neural network (DNN) dan self-feature extraction untuk mengatasi keberadaan data invalid (outlier) dari sinyal E-nose dalam kasus deteksi SARS melalui bau keringat ketiak. Metode yang diusulkan dalam tugas deteksi outlier memiliki kinerja yang menjanjikan dengan nilai rata-rata balanced accuracy (BA) sebesar 90,4%. Hasil ini menunjukkan bahwa DNN dan ekstraksi fitur yang diusulkan mengungguli metode konvensional seperti support vector machine (SVM), naïve bayes (NB), k-nearest neighbor (k-NN), dan kombinasi euclidean dengan z-score. Lalu, penggunaan standard scaler juga dapat meningkatkan performa model secara signifikan.
Sistem yang diusulkan dapat diimplementasikan untuk deteksi outlier secara real-time pada E-nose dan meningkatkan kinerja deteksi pasien SARS.
================================================================================================
Electronic Nose (E-nose) is a tool for imitating the nose work system that is cheap, easy and accurate in the process of implementing machine learning methods. Some researchers use E-nose to detect the authenticity of beef mixed with pork and detect food quality. In addition, in the medical field, E-nose is also widely used for research, such as early detection of patients with diabetes, lung cancer and tuberculosis through exhalation, as well as detecting infectious respiratory diseases (SARS) through the smell of sweat in the armpits.
The main challenge of using E-nose is outliers or data that are outside the population limit. Outliers are extreme cases that are different from other observations and the presence of outliers can damage data quality. In the case of detection of diabetes and tuberculosis through exhalation, the smell of the food or drink consumed can interfere with the detection process. Similarly, the detection of infectious respiratory diseases through the smell of underarm sweat. The use of deodorants and perfumes as well as odors around the sample room can impair data quality.
This study proposes an adaptive filter using a deep neural network (DNN) and self-feature extraction to overcome the presence of invalid data (outliers) from the E-nose signal in the case of SARS detection through the smell of underarm sweat. The method proposed in the outlier detection task has a promising performance with an average balanced accuracy (BA) of 90.4%. These results indicate that the proposed DNN and feature extraction outperform conventional methods such as support vector machine (SVM), naïve bayes (NB), k-nearest neighbor (k-NN), and the combination of euclidean with z-score. Also, using a standard scaler significantly improved the model performance.
The proposed system can be implemented for real-time outlier detection on E-nose and improve detection performance of SARS patients.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Artificial Olfaction, Deep Neural Network, Electronic Nose, Outlier Detection, Signal Processing, Deteksi Outlier, Pemrosesan Sinyal
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5102.9 Signal processing.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55101-(S2) Master Thesis
Depositing User:	Doni Putra Purbawa
Date Deposited:	07 Jul 2022 08:37
Last Modified:	01 Nov 2022 04:34
URI:	http://repository.its.ac.id/id/eprint/94967

Actions (login required)

View Item