ANALYSIS OF LONG SHORT – TERM MEMORY (LSTM) PARAMETERS IN PREDICTING IHSG

Resumo

For investors looking to enhance the value of their financial assets, stock investment is a popular choice. A Long Short-Term Memory (LSTM) model will be used to forecast the movement of the Indonesia Composite Index (IHSG) in the domestic capital market. This research focuses on key parameters of the LSTM model, such as sliding window size, the number of epochs, the learning rate, and the type of optimizer. There are four configurations that were tested. First, the sliding window size was varied while keeping other parameters constant. Second, while maintaining the other parameters, the number of epochs was modified. Third, while keeping the remaining parameters unchanged, the learning rate was adjusted. Lastly, while holding the other parameters constant, different optimizers were tested. The dataset is divided into two periods, such as: pre-pandemic and during the pandemic. The dataset is segmented into training and testing sets for every period. During the pre-pandemic period, the best-performing parameters included a sliding window size of 20, training over 40 epochs with a learning rate of 0.001, and the Adam optimizer, resulting in an RMSE of 7.2218. The best results during the pandemic period were obtained with parameters consisting of a sliding window size of 5, 10 epochs, a learning rate of 0.001, and the Adam optimizer, resulting in an RMSE of 1.727. These parameter combinations demonstrated the highest predictive performance for IHSG. BAHASA INDONESIA ABSTRACT: Untuk para investor yang ingin meningkatkan nilai aset keuangan mereka, investasi saham adalah pilihan populer. Sebuah model Long Short-Term Memory (LSTM) akan digunakan untuk memprediksi harga Indeks Harga Saham Gabungan (IHSG) di pasar modal Indonesia. Penelitian ini memfokuskan pada parameter kunci dari model LSTM, seperti ukuran sliding windows, jumlah epoch, learning rate, dan jenis optimizer. Ada empat konfigurasi yang diuji. Pertama, ukuran sliding windows divariasikan sementara parameter lainnya tetap konstan. Kedua, jumlah epoch dimodifikasi dengan tetap mempertahankan parameter lainnya. Ketiga, learning rate divariasikan dengan parameter lainnya tetap tidak berubah. Terakhir, berbagai optimizer diuji dengan parameter lainnya tetap konstan. Dataset ini dibagi menjadi dua periode, yaitu sebelum pandemi dan selama pandemi. Data dibagi menjadi set pelatihan dan pengujian untuk setiap periode. Parameter optimal untuk periode sebelum pandemi adalah ukuran sliding windows 20, 40 epoch, learning rate 0,001, dan optimizer Adam, menghasilkan Root Mean Squared Error (RMSE) sebesar 7,2218. Selama pandemi, parameter terbaik adalah ukuran sliding windows 5, 10 epoch, learning rate 0,001, dan optimizer Adam, dengan RMSE sebesar 1,727. Kombinasi parameter ini menunjukkan kinerja prediksi tertinggi untuk IHSG.

Dados da obra

Tipo

Artigo

Periódico

JOHME: Journal of Holistic Mathematics Education, v. 9, n. 2

Editora

Universitas Pelita Harapan

Ano

2025

Idioma

en

Palavras-chave

IHSG; LSTM; prediction; parameter; stock; prediksi

DOI

10.19166/johme.v9i2.10220

ISSN

2598-6759

Acesso

Acesso aberto

Fontes

DOAJ — Directory of Open Access Journals

Na Biblioteca desde

01/10/2026 · verificada em 03/10/2026