PREDIKSI HARGA SAHAM PT BANK MANDIRI (PERSERO) TBK PERIODE 2019–2024 MENGGUNAKAN METODE LONG SHORT-TERM MEMORY (LSTM) BERBASIS SISTEM WEB
Kata Kunci:
BMRI, Deep Learning, Long Short-Term Memory, Prediksi Harga Saham, Sistem Berbasis WebAbstrak
Pasar modal merupakan indikator penting perekonomian nasional, di mana volatilitas harga saham menuntut ketersediaan metode prediksi yang presisi untuk mendukung pengambilan keputusan investasi yang rasional. Penelitian ini bertujuan merancang, membangun, dan mengevaluasi model prediksi harga saham PT Bank Mandiri (Persero) Tbk (BMRI) periode 2019–2024 menggunakan arsitektur Long Short-Term Memory (LSTM), serta mengintegrasikannya ke dalam sistem berbasis web untuk mendukung reproduksibilitas dan aksesibilitas hasil penelitian. Data yang digunakan adalah data sekunder deret waktu univariat berupa harga penutupan harian (closing price) yang diperoleh dari Yahoo Finance. Metode penelitian mencakup prapemrosesan data (Min-Max Scaling), pembentukan data sekuensial dengan teknik sliding window, pembagian data secara kronologis dengan rasio 80:20, serta pengujian beberapa skenario hyperparameter. Hasil penelitian menunjukkan bahwa konfigurasi model terbaik dicapai pada arsitektur 2 layer LSTM dengan 50 neuron per layer, look-back window 30 hari, dropout rate 0,2, dan 50 epoch pelatihan, dengan optimizer Adam dan fungsi kerugian Mean Squared Error. Model tersebut menghasilkan nilai Root Mean Square Error (RMSE) sebesar 173,17, Mean Absolute Error (MAE) sebesar 139,09, dan Mean Absolute Percentage Error (MAPE) sebesar 2,13% yang termasuk kategori sangat akurat (excellent). Hasil ini membuktikan bahwa arsitektur LSTM yang diimplementasikan pada sistem berbasis web efektif dan andal dalam memprediksi pola pergerakan harga saham BMRI secara akurat, konsisten, dan dapat direproduksi.
The capital market is a key indicator of national economic performance, where volatile stock price movements require precise prediction methods to support rational investment decision-making. This study aims to design, build, and evaluate a stock price prediction model for PT Bank Mandiri (Persero) Tbk (BMRI) for the 2019–2024 period using a Long Short-Term Memory (LSTM) architecture, and to integrate it into a web-based system to support the reproducibility and accessibility of the research results. The data used is univariate time series secondary data in the form of daily closing prices obtained from Yahoo Finance. The research method includes data preprocessing (Min-Max Scaling), sequential data formation using a sliding window technique, chronological data splitting with an 80:20 ratio, and testing of several hyperparameter scenarios. The results show that the best model configuration is achieved with a 2-layer LSTM architecture of 50 neurons per layer, a 30-day look-back window, a 0.2 dropout rate, and 50 training epochs, using the Adam optimizer and Mean Squared Error loss function. The model produced a Root Mean Square Error (RMSE) of 173.17, a Mean Absolute Error (MAE) of 139.09, and a Mean Absolute Percentage Error (MAPE) of 2.13%, classified as an excellent accuracy category. These results demonstrate that the LSTM architecture implemented within a web-based system is effective and reliable for predicting BMRI stock price movements accurately, consistently, and reproducibly.




