<![CDATA[Stock price forecasting remains challenging because financial time-series data exhibit dynamic and nonlinear price movements. Although recent studies increasingly employ hybrid and attention-based deep learning architectures, the empirical evaluation of selected training and regularization hyperparameters in standalone LSTM models for individual Indonesian equities remains limited. This study aims to develop and evaluate a Long Short-Term Memory (LSTM) model with systematic hyperparameter tuning for ANTM stock price forecasting. A quantitative time-series approach was applied to 3,051 daily observations of ANTM stock prices obtained from Yahoo Finance from January 1, 2014, to June 6, 2026, using open, high, low, and close (OHLC) variables. The model consists of three LSTM layers with dropout and a dense output layer, while Grid Search was used to evaluate combinations of batch size and dropout rate. Model performance was assessed using MAPE, RMSE, and R². The best configuration, with a batch size of 32 and dropout rate of 0.3, achieved a validation loss of 0.000617. On the testing data, the model obtained a MAPE of 6.62%, RMSE of 257.40, and R² of 0.9351. The resulting model was further used to generate a 12-month point forecast, which indicated a gradual downward trajectory from Rp3,117.31 in June 2026 to Rp2,820.05 in May 2027. The findings provide empirical evidence on the performance of a systematically tuned standalone LSTM for ANTM stock price forecasting and highlight the importance of configuration selection in financial time-series modelling.]]>
