Implementasi Attention – Long Short Term Memory dalam Memprediksi Harga Saham Berdasarkan Data Historis dan Analisis Sentimen

Abstract

The capital market plays an important role in the economy by providing opportunities for people to invest, especially in stock instruments. Stock price prediction is crucial for various stakeholders, such as investors, consultants, and governments, to manage portfolios, gain profits, and maintain financial stability. Due to the rapid fluctuation of stock prices and their non-linear nature, an accurate prediction model is needed. This study uses the Attention - Long Short Term Memory (LSTM) algorithm, which is known to be effective for time series data, to predict stock prices. In addition, sentiment analysis is carried out using the lexicon-based VADER method, which has been proven to be more accurate for text on social media than other methods such as Naïve Bayes and SVM. The integration of the Attention mechanism with LSTM is expected to enhance the accuracy of stock price predictions, thereby providing more reliable information for investors in their decision-making processes. The best model generated from this study achieved a Mean Squared Error (MSE) of 0.041, a Mean Absolute Error (MAE) of 0.180, and a Root Mean Squared Error (RMSE) of 0.020 on the test data. Meanwhile, the model achieved an MAE of 0.028, an MSE of 0.001, and an RMSE of 0.035 on the training data.