gradDescentR 2.0 : IMPLEMENTASI METODE GRADIENT DESCENT DAN VARIASINYA DALAM R PACKAGE

Handian, Dendi (2017) gradDescentR 2.0 : IMPLEMENTASI METODE GRADIENT DESCENT DAN VARIASINYA DALAM R PACKAGE. Other thesis, Universitas Pendidikan Indonesia.

[img]
Preview
Text
S_IKOM_1200678_Title.pdf

Download (127kB) | Preview
[img]
Preview
Text
S_IKOM_1200678_Abstract.pdf

Download (120kB) | Preview
[img]
Preview
Text
S_IKOM_1200678_Table_of_content.pdf

Download (289kB) | Preview
[img]
Preview
Text
S_IKOM_1200678_Chapter1.pdf

Download (116kB) | Preview
[img] Text
S_IKOM_1200678_Chapter2.pdf
Restricted to Repository staff only

Download (1MB)
[img]
Preview
Text
S_IKOM_1200678_Chapter3.pdf

Download (583kB) | Preview
[img] Text
S_IKOM_1200678_Chapter4.pdf
Restricted to Repository staff only

Download (1MB)
[img]
Preview
Text
S_IKOM_1200678_Chapter5.pdf

Download (102kB) | Preview
[img]
Preview
Text
S_IKOM_1200678_Bibliography.pdf

Download (295kB) | Preview
Official URL: http://repository.upi.edu

Abstract

Machine learning merupakan cabang ilmu komputer yang berfokus pada algoritma yang bisa belajar dari data. Regresi merupakan tugas dari supervised learning yang digunakan untuk memprediksi nilai riil berdasarkan variabel predictor. Prediksi yang akurat pada regresi ditandai dengan cost function yang minimum. Metode gradient descent (GD) digunakan untuk mencari nilai minimum lokal dari suatu fungsi. Sepanjang waktu GD terus berkembang, sehingga banyak dikembangkan berbagai variasi GD seperti Mini-Batch Gradient Descent (MBGD), Stochastic Gradient Descent (SGD), Stochastic Average Gradient Descent (SAGD), Momentum Gradient Descent (MGD), Accelerated Gradient Descent (AGD), Adagrad, Adadelta, RMSprop dan Adam. Penelitian ini berfokus pada pengembangan R package bernama gradDescentR, yang mengimplementasikan metode GD dan variasinya untuk melakukan prediksi pada tugas regresi. R merupakan bahasa pemrograman yang populer dalam kasus analisis data, statistik dan machine learning. Untuk menguji R package ini dilakukan eksperimen dan simulasi untuk mencari atau memprediksi nilai faktor kompresibilitas gas CO2 berdasarkan parameter tekanan dan suhu yang didapatkan. Penelitian tentang implementasi metode berbasis GD untuk memprediksi faktor kompresibilitas gas CO2 menggunakan R sudah pernah dilakukan sebelumnya. Namun pada penelitian tersebut, metode yang diimplementasikan hanya ada empat yaitu GD, MBGD, SGD dan SAGD. Penelitian ini akan melanjutkan penelitian tersebut dengan menambahkan metode lain yaitu MGD, AGD, Adagrad, Adadelta, RMSprop dan Adam. Berdasarkan hasil penelitian yang dilakukan, gradDescentR berhasil dikembangkan dan sudah terpublikasi pada Comprehensive R Archive Network (CRAN). Eksperimen dan simulasi R Package pada studi kasus faktor kompresibilitas gas CO2 telah selesai dilakukan dengan hasil rata-rata untuk nilai RMSE sebesar 0.008521815 dan waktu eksekusi sebesar 0.14266663 detik.;--- Machine learning is a field of computer science that focus on algorithms that able to learn from data. Regression is one of the supervised learning tasks used to predict a real value (dependent variable) based on predictor value (independent variable). An accurate prediction indicated by a low cost function. Gradient Descent (GD) method used to find local minimal of an objective function. Until these days, GD had been developed into various algorithms, such as Mini-Batch Gradient Descent (MBGD), Stochastic Gradient Descent (SGD), Stochastic Average Gradient Descent (SAGD), Momentum Gradient Descent (MGD), Accelerated Gradient Descent (AGD), Adagrad, Adadelta, RMSprop and Adam. This research focuses on developing an R Package named gradDescentR, which it is implementing GD method and its variants to perform prediction on regression task. R is a popular programming language used for data analysis, statistics, and machine learning. To make use of this R package, we propose a case study of CO2 gas compressibility factor or simply z-factor, to predict its value that depends on acquired temperature and pressure. The research about predicting z-factor value by using GD method in R has been done. However, in the research, the implemented method only has four, which is GD, MBGD, SGD and SAGD. This research will continue its work by adding more method, such as MGD, AGD, Adagrad, Adadelta, RMSprop and Adam. As the result of this research, gradDescentR have been built and published in Comprehensive R Archive Network (CRAN). The experiment and simulating the R Package on CO2 Gas Compressibility Factor is done with the average result for prediction Root-Mean-Square-Error (RMSE) has a value of 0.008521815 and the estimated built time of 0.14266663 seconds.

Item Type: Thesis (Other)
Additional Information: No.panggil : S IKOM HAN g-2017; Pembimbing : I.Lala Septemriza, II.Rani Megasari.
Uncontrolled Keywords: machine learning, regresi, gradient descent, R, R Package, CRAN, faktor kompresibilitas gas, machine learning, regression, gradient descent, R, R package, CRAN, gas compressibility factor.
Subjects: Q Science > Q Science (General)
Divisions: Fakultas Pendidikan Matematika dan Ilmu Pengetahuan Alam > Program Studi Ilmu Komputer
Depositing User: Mr mhsinf 2017
Date Deposited: 10 Jan 2018 07:17
Last Modified: 10 Jan 2018 07:17
URI: http://repository.upi.edu/id/eprint/28457

Actions (login required)

View Item View Item