<![CDATA[Nickel is a strategic commodity that plays an important role in the global industry, particularly as a key material for electric vehicle batteries. PT Vale Indonesia Tbk (INCO), a nickel mining company listed on the Indonesia Stock Exchange, experiences stock price fluctuations driven by technical and macroeconomic factors. This study aims to develop a forecasting model for INCO's stock price using Bidirectional Long Short-Term Memory (Bi-LSTM) optimized with a Genetic Algorithm (GA) and to identify the most influential predictor variables using Shapley Additive Explanations (SHAP). Monthly data from January 2007 to December 2025 include INCO's stock price, nickel price, exchange rate, inflation, and the BI-Rate. A forward selection procedure was applied to determine the best predictor combination, after which the model was trained using the Adam Optimizer with GA-optimized hyperparameters and evaluated using Mean Absolute Percentage Error (MAPE). The best model was obtained from the combination of historical stock price, nickel price, and the BI-Rate, with optimal hyperparameters of 150 epochs, a batch size of 20, 150 neurons, a learning rate of 0.002844, and a dropout of 0.0502, producing a MAPE of 8.751%. SHAP results indicate that historical stock price and nickel price contribute the most to the prediction, whereas the BI-Rate contributes relatively less. This study concludes that combining Bi-LSTM, GA, and predictor variable selection improves the forecasting accuracy of INCO's stock price and provides useful insights for long-term investment decision-making in the nickel mining sector. Keywords: Bidirectional Long Short Term Memory, Forecasting, Genetic Algorithm, Macroeconomics, Stocks]]>
Copyrights © 2026
