2Deep Model Prediksi Berbasis Weighting Average Untuk Time Series Data

Abstract

In time series data analysis, the need for accurate and efficient predictive models is becoming increasingly urgent as data complexity rises. This study proposes the 2Deep Model, a hybrid approach that combines Bidirectional Long Short-Term Memory (Bi-LSTM) and Stacked LSTM, utilizing the Weighting Average technique to optimize predictions. This method was chosen for its potential in handling long-term dependencies and temporal complexity in data. Experiments were conducted on five datasets: ETTh1, ETTh2, ETTm1, ETTm2, and AQI Shanghai. The results show that the proposed model achieves low Mean Squared Error (MSE) and Mean Absolute Error (MAE) values on the first four datasets, with an average MSE of 0.0289 and an MAE of 0.0971, along with a relatively high R-squared (R²) value. However, for the AQI Shanghai dataset, the model's performance declined, with higher MSE and MAE values and a lower R². These findings indicate that the 2Deep Model holds significant potential for time series data prediction applications, although there is room for improvement when dealing with more diverse datasets. Future research suggestions include further model optimization and exploring other hybrid methods to enhance model generalization.