PREDIKSI HARGA SAHAM BANK MANDIRI BERDASARKAN DATA BMRI HISTORICAL STOCK PRICE

Abstract

Stock price movements exhibit dynamic and highly fluctuating characteristics, making accurate prediction challenging when using conventional approaches. Therefore, this study aims to develop and evaluate a stock price prediction model for PT Bank Mandiri (Persero) Tbk using a deep learning approach based on Long Short-Term Memory (LSTM). The LSTM model is applied for time series forecasting by utilizing historical stock price data, with a primary focus on the closing price as the target variable. The historical stock price data of Bank Mandiri undergo preprocessing stages, including data cleaning and normalization using the Min–Max Scaling method, to align data scales and improve training stability and convergence. The proposed LSTM architecture consists of two LSTM layers with dropout mechanisms for regularization, followed by fully connected layers to generate stock price predictions. The model is trained using the Adam optimizer with Mean Squared Error (MSE) as the loss function. Model performance is evaluated using the Mean Absolute Percentage Error (MAPE) metric on the testing dataset. The experimental results show that the LSTM model achieves a MAPE value of 2.7572%, indicating a very high prediction accuracy. Furthermore, the future forecasting results suggest a relatively stable stock price movement with a gradual upward trend in the short term. Based on the findings, it can be concluded that the Long Short-Term Memory (LSTM) method is effective for predicting Bank Mandiri’s stock prices and has strong potential as a data-driven decision support tool for investment analysis, although it should be complemented with fundamental analysis and external market factors