Time Series Modeling of Stock Price Using CNN-BiLSTM with Attention Mechanism

Abstrak

Indonesia's capital market has experienced rapid development in recent years, marked by an increase in transaction value, the number of investors, and market capitalization. One of the sectors that has garnered attention is the telecommunications industry, which is rapidly growing alongside the increasing number of internet users and the public's demand for more advanced telecommunications services. PT Indosat Ooredoo Hutchison, as one of the leading telecommunications companies in Indonesia, has become an attractive investment choice for investors. However, the stock market is known for its fluctuating and irregular nature. Stock data has complex characteristics such as large data volume, ambiguous information, and non-linearity. Therefore, it is important for investors to understand stock price movements before making investments in order to reduce the risk of significant losses. One method that can be used to address that risk is by forecasting stock prices. Time series forecasting is a prediction about future values based on historical data. Statistical methods in forecasting allow for the identification of patterns and trends in historical data, as well as modeling the relationships between variables over time. One of the techniques that is becoming increasingly popular in forecasting is deep learning. In this study, a combination of \textit{Convolutional Neural Network} (CNN) and \textit{Bidirectional Long Short-Term Memory} (BiLSTM) with an attention mechanism is used. CNN excels at extracting data features, while BiLSTM is better at handling data with long time ranges. The addition of the attention mechanism allows the model to assign different weights to data features, enabling it to focus on the most relevant information. The combination of these three elements (CNN-BiLSTM with an attention mechanism) has the potential to yield higher prediction accuracy. To measure the accuracy of the forecasts, this study uses evaluation metrics such as Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Root Mean Squared Error (RMSE), and R-squared (R²). The research results indicate that the CNN-BiLSTM model with an attention mechanism has proven to be the most superior model compared to other models in forecasting the stock price of PT Indosat Ooredoo Hutchison.

Kata kunci

• Time Series Forecasting
• CNN
• BiLSTM
• Attention Mechanism
• stock prices