Santoso, Reinato Teguh and Adharul Muttaqin, S.T., M.T. and Dr. Eng. Panca Mudjirahadrjo,, S.T., M.T. (2023) Ekstraksi Ciri Berdasarkan Karakteristik Dinamis Sinyal Multisensor Menggunakan Linear Discriminant Analysis. Sarjana thesis, Universitas Brawijaya.
Abstract
Penelitian ini bertujuan untuk mengimplementasikan ekstraksi ciri berdasarkan pada karakteristik dinamis sinyal multisensor. Data yang digunakan pada penelitian ini merupakan hasil pengukuran aroma 6 jenis mint yang dilakukan di Botanical Institute of Karlsruhe Institute of Technology (KIT) Jerman. Data terdiri atas respons frekuensi dari sensor QCM dengan 12 bahan lapisan nanomaterial yang berbeda, berjumlah 28.746 data, dengan waktu pengambilan sampel rata-rata pada tiap jenis mint adalah 175,52 menit. Pada data kemudian dilakukan tahap preprocessing untuk mengatasi nilai tidak beraturan dan melakukan standarisasi pada data. Selanjutnya, dilakukan 2 tahap pengujian yaitu pengujian performa ekstraksi ciri dan pengujian performa klasifikasi. Ekstraksi ciri dilakukan menggunakan 2 jenis data yang berbeda, yaitu data respons frekuensi dan data karakteristik dinamis hasil PLR. Ekstraksi ciri menggunakan data respons frekuensi menghasilkan 5 fitur tereduksi dengan persentase eigenvalue LD1 56,75%, LD2 19,03%, LD3 14,09%, LD4 7,32%, dan LD5 2,81%. Ekstraksi ciri menggunakan data karakteristik dinamis juga menghasilkan 5 fitur tereduksi dengan persentase eigenvalue LD1 61,12%, LD2 27,70%, LD3 7,67%, LD4 2,38%, dan LD5 1,12%. Pengujian performa klasifikasi menggunakan 3 model machine learning berbeda yaitu k-NN, SVM, dan DTC. Klasifikasi menggunakan data respons frekuensi sebelum LDA menghasilkan rata-rata akurasi 82,08% dan waktu komputasi 13,326 detik, setelah LDA menghasilkan rata-rata akurasi 96,48% dan waktu komputasi 1,398 detik, menggunakan data PLR sebelum LDA menghasilkan rata-rata akurasi 85,07% dan waktu komputasi 0,263 detik, setelah LDA menghasilkan rata-rata akurasi 91,67% dan waktu komputasi 0,071 detik. Hasil klasifikasi menunjukkan bahwa ekstraksi ciri menggunakan LDA dapat meningkatkan nilai akurasi klasifikasi meskipun dimensi data tereduksi.
English Abstract
This research aims to implement feature extraction based on the dynamic characteristics of multisensor signals. The data used in this study were obtained from measurements of 6 types of mint aromas at the Botanical Institute of Karlsruhe Institute of Technology (KIT) in Germany. The dataset consists of frequency responses from a QCM sensor with 12 different nanomaterial layer materials, involving 28,746 data points, with an average sampling time for each type of mint at 175.52 minutes. The data then underwent a preprocessing stage to handle irregular values and standardize the data. Subsequently, two testing stages were conducted: feature extraction performance testing and classification performance testing. Feature extraction was performed using two different types of data: frequency response data and dynamic characteristic data obtained from PLR. Feature extraction using frequency response data resulted in 5 reduced features with eigenvalue percentages for LD1 at 56.75%, LD2 at 19.03%, LD3 at 14.09%, LD4 at 7.32%, and LD5 at 2.81%. On the other hand, feature extraction using dynamic characteristic data also yielded 5 reduced features with eigenvalue percentages for LD1 at 61.12%, LD2 at 27.70%, LD3 at 7.67%, LD4 at 2.38%, and LD5 at 1.12%. For classification performance testing, three different machine learning models were used: k-NN, SVM, and DTC. Using frequency response data before LDA (Linear Discriminant Analysis), the average accuracy was 82.08% with a computation time of 13.326 seconds. After LDA, the average accuracy increased to 96.48% with a computation time of 1.398 seconds. Using PLR data before LDA, the average accuracy was 85.07% with a computation time of 0.263 seconds, and after LDA, the average accuracy increased to 91.67% with a computation time of 0.071 seconds. The classification results indicate that feature extraction using LDA can improve the classification accuracy even with reduced data dimensions.
| Item Type: | Thesis (Sarjana) |
|---|---|
| Identification Number: | 0523070364 |
| Uncontrolled Keywords: | Ekstraksi Ciri, Multisensor, Quartz Crystal Microbalance, Linear Discriminant Analysis |
| Subjects: | 600 Technology (Applied sciences) > 621 Applied physics > 621.3 Electrical, magnetic, optical, communications, computer engineering; electronics, lighting > 621.38 Electronics, communications engineering > 621.381 Electronics |
| Divisions: | Fakultas Teknik > Teknik Elektro |
| Depositing User: | Endang Susworini |
| Date Deposited: | 09 Oct 2023 01:44 |
| Last Modified: | 09 Oct 2023 01:44 |
| URI: | http://repository.ub.ac.id/id/eprint/203696 |
![[thumbnail of DALAM MASA EMBARGO]](http://repository.ub.ac.id/style/images/fileicons/text.png) |
Text (DALAM MASA EMBARGO)
Reinato Teguh Santoso.pdf Restricted to Registered users only until 31 December 2025. Download (6MB) |
Actions (login required)
 |
View Item |