Deep Learning Approach to Pharmaceutical Stock Forecasting using LSTM Architecture

Abstract

Stock price forecasting in the pharmaceutical sector is challenging due to high volatility and nonlinear temporal patterns. Conventional statistical models often fail to capture long-term dependencies in financial time series. This study proposes an optimized Long Short-Term Memory (LSTM) architecture to forecast the closing stock prices of PT Kalbe Farma Tbk (KLBF.JK). Historical daily stock data from 2020 to 2024 were collected from Yahoo Finance and preprocessed using Min–Max normalization. In this study, we evaluate several LSTM by varying epochs and batch sizes to identify the optimal model. Experimental results show that the proposed LSTM model achieved the lowest Root Mean Square Error (RMSE) of 25.1406 using 100 epochs and a batch size of 5. The configured LSTM demonstrating superior predictive performance in capturing stock price dynamics. The findings confirm that the optimized LSTM architecture is effective for pharmaceutical stock forecasting and can support data-driven investment decision-making.