Pradani, Humaira Nur (2022) Ekstraksi Fitur Dengan Polynomial Detrended Fluctuation Analysis (P-Dfa) Untuk Pengembangan Model Human Activity Recognition (Har). Masters thesis, Institut Teknologi Sepuluh Nopember.
|
Text
6026201014-Master_Thesis.pdf Restricted to Repository staff only Download (4MB) | Request a copy |
Abstract
Human Activity Recognition (HAR) merupakan salah satu topik dalam bioinformatika yang berfokus untuk mengenali aktivitas manusia sehingga dapat membantu dalam membuat keputusan tertentu atau menghasilkan peringatan dini. Seiring dengan berkembangnya MEMS (Micro Electro Mechanical Systems), akuisisi data HAR dapat dilakukan dengan lebih efisien menggunakan sensor inersia yang terdapat pada berbagai perangkat pintar. Pengolahan data sensor inersia untuk HAR membutuhkan serangkaian proses, dimana salah satu tahapan paling krusial adalah tahap ekstraksi fitur. Kombinasi fitur yang masuk pada model klasifikasi dapat mempengaruhi performa model yang bisa diukur dari nilai akurasi. Sehingga, diperlukan fitur-fitur yang optimal untuk menciptakan model HAR yang baik. Penelitian ini mengusulkan pembuatan model klasifikasi HAR dengan ekstraksi fitur menggunakan Polynomial Detrended Fluctuation Analysis (P-DFA) yang merupakan pengembangan dari metode Detrended Fluctuation Analysis (DFA). Pada penelitian ini, Fitur P-DFA diujicobakan dengan berbagai kombinasi fitur untuk menjadi fitur masukan pada model HAR menggunakan algoritma klasifikasi tertentu. Dari uji coba yang dilakukan pada dataset MotionSense, secara umum penambahan fitur P-DFA mampu meningkatkan akurasi model HAR. Model HAR terbaik yang dihasilkan dengan algoritma klasifikasi Gradient Boosting Decision Tree dan kombinasi fitur P-DFA, Time Domain, serta Frequency Domain mampu memiliki performa yang mengungguli metode state-of-the-art. Model ini mampu menghasilkan nilai akurasi sebesar 97,4%, presisi sebesar 96,4%, serta recall sebesar 96,5%. Di samping itu, penambahan ekstraksi fitur P-DFA memiliki konsekuensi perlambatan runtime saat implementasi model HAR hanya sebesar ± 0,28 detik. Metode yang diusulkan juga mampu menghasilkan akurasi yang baik pada dataset lain, yakni dataset MHEALTH dengan nilai akurasi sebesar 90,7%.
==============================================================================================================================
Human Activity Recognition (HAR) is one of the topics in bioinformatics that focuses on recognizing human activities so that it can assist in making certain decisions or early warnings. Along with the development of MEMS (Micro Electro Mechanical Systems), HAR data acquisition can be carried out more efficiently using inertial sensors found in various devices. Processing inertial sensor data for HAR requires a process, where one of the most important stages is the feature extraction stage. The combination of features included in the classification model can affect the performance of the model which can be measured from the accuracy value. Thus, optimal features are needed to create a good HAR model. This study proposes a HAR classification model with feature extraction using Polynomial Detrended Fluctuation Analysis (P-DFA), which is the development of the Detrended Fluctuation Analysis (DFA) method. The P-DFA feature is tested with various combinations of features to become an input feature in the HAR model using certain classification algorithms. From the experiments conducted on the MotionSense dataset, in general, the addition of the P-DFA feature is able to increase the accuracy of the HAR model. The best HAR model generated by the Gradient Boosting Decision Tree classification and the combination of P-DFA, Time Domain, and Frequency Domain features is able to have a performance that outperforms state-of-the-art methods. This model is able to produce an accuracy value of 97.4%, a precision of 96.4%, and a recall of 96.5%. In addition, the addition of P-DFA feature extraction has a runtime load when the implementation of the HAR model is only ± 0.28 seconds. The proposed method is also able to produce good accuracy on other datasets, namely the MHEALTH dataset with an accuracy value of 90.7%.
| Item Type: | Thesis (Masters) |
|---|---|
| Additional Information: | RTSI 006.31 Pra e-1 2022 |
| Uncontrolled Keywords: | Detrended Fluctuation Analysis (DFA), Human Activity Recognition (HAR), Regresi Polinomial, Sensor Inersia. Detrended Fluctuation Analysis (DFA), Human Activity Recognition (HAR), Inertial Sensor, Polynomial Regression. |
| Subjects: | T Technology > T Technology (General) > T58.6 Management information systems |
| Divisions: | Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information System > 59101-(S2) Master Thesis |
| Depositing User: | Mr. Marsudiyana - |
| Date Deposited: | 07 Jul 2026 03:28 |
| Last Modified: | 07 Jul 2026 03:28 |
| URI: | http://repository.its.ac.id/id/eprint/134406 |
Actions (login required)
![]() |
View Item |
