Jaringan Syaraf Backpropagationdan Algoritma Genetika Untuk Meramalkan Indeks Barga Saham Gabungan

Halim, Yunus Abdul (2005) Jaringan Syaraf Backpropagationdan Algoritma Genetika Untuk Meramalkan Indeks Barga Saham Gabungan. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 5199201018-Master_thesis.pdf Restricted to Repository staff only Download (23MB)

Abstract

Peramalan dilakukan untuk mengestimasi suatu perilaku data berdasarkan analisis dan pengolahan data historis. Data historis mempunyai keterkaitan terhadap analisis karakteristik dan pola data. Jaringan syaraf tiruan backpropagation merupakan metode yang efisien untuk meramalkan pasar modal dan indeks harga saham. Jaringan syaraf backpropagation menggunakan perubahan bobot secara heuristik dalam pencarian nilai optimum. Di lain pihak, algoritma genetika mampu memberikan nilai bobot secara acak dan dengan cepat mampu menemukan nilai optimum. Perubahan bobot secara heuristik pada jaringan syaraf backpropagation disempurnakan dengan menggabungkan algoritma genetika ke dalam jaringan syaraf tiruan. Dalam penelitian ini diimplementasikan algoritma genetika untuk menyempurnakan jaringan syaraf backpropagation. Algoritma genetika digunakan untuk menentukan nilai bobot yang dijadikan nilai masukan oleh jaringan syaraf backpropagation. Data yang digunakan untuk pelatihan dan pengujian adalah data indeks harga saham gabungan (IHSG) pada Bursa Efek Jakarta. Uji coba dilakukan dengan menggunakan struktur jaringan syaraf multilayer perceptron. Pada proses pelatihan dilakukan uji coba dengan variasi jumlah lapisan masukan dan jumlah lapisan tersembunyi untuk menghasilkan model terbaik. Tolok ukur keberhasilan pengujian model dilakukan dengan menghitung mean square error (MSE), standar deviasi (SD), dan U-Theil. Perangkat lunak yang telah berhasil dibuat telah diuji coba untuk meramalkan indeks harga saham gabungan (IHSG) di Bursa Efek Jakarta. Data yang digunakan adalah data IHSG tahun 2004 dengan beberapa variasi pengujian untuk data tahun 2005. Hasil uji coba menunjukkan bahwa perangkat lunak penggabungan jaringan syaraf backpropagation dan algoritma genetika mampu meningkatkan kecepatan jaringan syaraf tiruan backpropagation dalam mencari nilai optimum sebesar 82%. Pengaruh penambahan jumlah lapisan tersembunyi mengakibatkan nilai MSE jaringan syaraf backpropagation semakin kecil dan tidak berpengaruh terhadap penggabungan jaringan syaraf backpropagation dan algoritma genetika. Penggabungan jaringan syaraf backpropagation dan algoritma genetika mampu menghasilkan nilai U-Theil (lebih kecil dari 0,8), MSE, dan standar deviasi lebih kecil dibandingkan dengan hasil yang diberikan hanya dengan menggunakan jaringan syaraf backpropagation.
====================================================================================================================================
A forecasting process is usually involved in estimating data patterns based on the analysis of historical data that are strongly related to the characteristics of the data. An artificial backpropagation neural network (BPNN) is known as one of the efficient methods for forecasting the Stock Exchange Composite Index (SECI). In this context, the BPNN heuristically uses changes in network weights to obtain an optimal solution. On the other hand, the genetic algorithm (GA) has the capability of providing initial random weights that can then be further processed in a relatively short computing time to obtain the optimal solution. Therefore, a merging algorithm that involves the GA in the BPNN for forecasting the SECI is needed to improve the performance of the forecasting process. This research implements a merging algorithm that combines the GA and BPNN, which is called GA-BPNN, in performing the forecasting process of the SECI. In this approach, GA is used to determine the initial weights of the BPNN. For the purpose of evaluating the performance of GA-BPNN, a multilayer perceptron neural network structure is used. During experimentation, a number of different structures, each with different numbers of input and hidden layers, are tested to evaluate the performance of the BPNN. The performance evaluation of the forecasting process is conducted by calculating the mean square error (MSE), standard deviation (STD), and U-Theil of the forecasting results compared to the actual data. The GA-BPNN has been tested using the SECI obtained from the Jakarta Stock Exchange. The SECI data used include data for all months of 2004 and up to July 2005. The entire data for 2004 together with the actual data for January 2005 are then used as training data for forecasting the SECI for February 2005. The same experimental method is used for subsequent forecasting processes until the SECI for July 2005 is forecast. The experimental results show that GA-BPNN is capable of reducing, on average, as much as 82% of the total computing time required by BPNN to reach its optimal value. Moreover, the values of MSE, STD, and U-Theil (less than 0.8) produced by GA-BPNN are lower than those produced by BPNN. Finally, although increasing the number of hidden layers in BPNN can reduce the MSE, it does not affect the performance of GA-BPNN.

Item Type:	Thesis (Masters)
Additional Information:	RTIf 006.32 Hal j
Subjects:	Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Information and Communication Technology > Informatics > 55101-(S2) Master Thesis
Depositing User:	magang .
Date Deposited:	22 Sep 2026 01:49
Last Modified:	22 Sep 2026 01:49
URI:	http://repository.its.ac.id/id/eprint/144766

Actions (login required)

View Item