Search for collections on Undip Repository

OPTIMASI MODEL LONG SHORT TERM MEMORY (LSTM) UNTUK SISTEM INFORMASI DETEKSI AKTIVITAS LANJUT USIA

SEPTIADI, Jaka and Warsito, Budi and Wibowo, Adi (2020) OPTIMASI MODEL LONG SHORT TERM MEMORY (LSTM) UNTUK SISTEM INFORMASI DETEKSI AKTIVITAS LANJUT USIA. Masters thesis, School of Postgraduate Studies.

[img] Text
COVER dll.pdf

Download (1MB)
[img] Text
13. BAB I.pdf

Download (107kB)
[img] Text
14. BAB II.pdf

Download (485kB)
[img] Text
15. BAB III.pdf
Restricted to Repository staff only

Download (811kB)
[img] Text
16. BAB IV.pdf
Restricted to Repository staff only

Download (1MB)
[img] Text
17. BAB V.pdf

Download (98kB)
[img] Text
18. DAFTAR PUSTAKA.pdf

Download (219kB)

Abstract

Salah satu ciri gejala demensia yang umum terjadi pada lanjut usia adalah tidak mengulangi kebiasaan aktivitas yang dilakukan dalam kesehariannya. Deteksi dini gejala demensia pada lanjut usia dapat dilakukan dengan mengidentifikasi kebiasaan aktivitas yang dilakukan di periode waktu tertentu dalam kesehariannya. Dalam penelitian ini dilakukan deteksi kebiasaan aktivitas lanjut usia berdasarkan data histori pemantauan aktivitas selama 19 hari. Model Long Short Term Memory (LSTM) menjadi salah satu model yang banyak digunakan untuk memprediksi aktivitas berdasarkan data histori, namun model LSTM masih memiliki kekurangan terkait lambatnya model dalam mencapai konvergensi. Dalam penelitian ini pengoptimalan dan penyesuaian parameter dilakukan untuk mengatasi permasalahan tersebut. Metode optimasi yang digunakan adalah Stochastic Gradient Descent (SGD), Adam, dan proses transisi optimasi Adam ke optimasi SGD, dan parameter yang disesuaikan yaitu penentuan nilai learning rate dan pemilihan fungsi aktivasi. Hasil pengujian model LSTM menunjukkan bahwa proses transisi optimasi Adam ke optimasi SGD lebih cepat dalam mencapai konvergensi dan menghasilkan generalisasi yang lebih baik pada epoch transisi 40 dengan nilai learning rate 0,001 untuk optimasi Adam dan 0,01 untuk optimasi SGD.
Kata kunci: Pengenalan Aktivitas Manusia, Long Short Term Memory (LSTM), Adam, Stochastic Gradient Descent (SGD)

One of the characteristics of dementia symptoms that commonly occur in the elderly people is that they do not repeat their daily activities. Early detection of dementia symptoms in the elderly people can be done by identifying their activity that are carried out in their daily lives. In this study, the activity of the elderly were detected based on human activity recognition historical data for 19 days. Long Short Term Memory (LSTM) model is one of the most widely used models for predicting activities based on historical data. However, LSTM model still has a drawback related to its slowness of the network in achieving convergence. In this research, optimization and determination of parameters are carried out to solve these problems. The optimization method used are Stochastic Gradient Descent (SGD), Adam, and switching from Adam to SGD during training, and the adjusted parameters are determining the learning rate and activation function. The results showed that switching from Adam to SGD during training is faster in achieving convergence and get better generalization in epoch transisition 40 with learning rate 0,001 for Adam and 0,01 for SGD after switching.
Keywords: Human Activity Recognition, Long Short Term Memory (LSTM), Adam, Stochastic Gradien Descent (SGD)

Item Type: Thesis (Masters)
Uncontrolled Keywords: Pengenalan Aktivitas Manusia, Long Short Term Memory (LSTM), Adam, Stochastic Gradient Descent (SGD)
Subjects: Sciences and Mathemathic
Divisions: Postgraduate Program > Master Program in Information System
Depositing User: ekana listianawati
Date Deposited: 26 Apr 2022 03:16
Last Modified: 26 Apr 2022 03:16
URI: https://eprints2.undip.ac.id/id/eprint/5976

Actions (login required)

View Item View Item