Prediksi Harga Saham Indonesia Menggunakan Transformer Encoder

Aulia, Fathimah (2023) Prediksi Harga Saham Indonesia Menggunakan Transformer Encoder. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 06111940000048-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2025. Download (2MB) | Request a copy

Abstract

Pasar saham adalah tempat dimana perusahaan menjual sahamnya kepada publik. Di Indonesia sendiri, pasar saham tercatat berkembang stabil dan menunjukkan kinerja positif. Dengan tingginya potensi tersebut, prediksi harga saham sangat diperlukan untuk mendapat keuntungan semaksimal mungkin, terutama bagi investor dan pelaku bisnis. Bagaimanapun, tugas prediksi saham merupakan hal yang tidak mudah karena pergerakan harga saham yang volatile. Pada penelitian ini diimplementasikan Transformer Encoder yang merupakan bagian dari model Transformer untuk memprediksi harga saham di Indonesia. Model Transformer digunakan karena arsitekturnya dapat beradaptasi dengan belajar dari data seri waktu yang non-linier dan kompleks. Dalam penelitian menggunakan data historis saham dari 1 Oktober 2005 hingga 1 April 2023, model dengan nilai error terendah ditemukan pada model yang dilatih dengan data gabungan saham-saham Indonesia. Rata-rata MSE (mean squared error) dari model tersebut adalah 0.720315. Uji coba penelitian juga dilakukan dengan memperkenalkan embedding waktu Time2Vec ke dalam model, tetapi hal ini hanya menghasilkan perbedaan error yang kecil. Hal ini mengindikasikan bahwa tambahan Time2Vec sebagai embedding waktu memiliki signifikansi yang rendah terhadap hasil untuk dataset yang digunakan.
=================================================================================================================================
The stock market is where companies sell their shares to the public. In Indonesia, the stock market has been growing steadily and showing positive performance. Given its high potential, stock price prediction is necessary to get the maximum profit possible, especially for investors and business person. However, predicting stock is not an easy task because stock price movements are volatile. This study implements Trannsformer-Encoder which is a part of the Transformer model to predict stock prices in Indonesia. The Transformer model is used because its architecture can adapt to learning from non-linear and complex time series data. In this study using stock historical data from October 1 2005 to April 1 2023, the model with the lowest error value was found from the model trained using combined data of Indonesian stocks. The average MSE (Mean Squared Error) of the model is 0.720315. Research trials were also carried out by introducing the Time2Vec into the model, but this only resulted in a small difference in error. This indicates that the additional embedding time using Time2Vec has a low significance on the result for the dataset used.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Transformer Encoder, stock price prediction, Indonesian stocks, prediksi harga saham, saham-saham di Indonesia
Subjects:	Q Science > QA Mathematics Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Mathematics > 44201-(S1) Undergraduate Thesis
Depositing User:	Fathimah Aulia
Date Deposited:	28 Dec 2023 02:04
Last Modified:	28 Dec 2023 02:04
URI:	http://repository.its.ac.id/id/eprint/101280

Actions (login required)

View Item