<![CDATA[This study aims to evaluate and compare the performance of several variants of Long Short-Term Memory (LSTM) based models in predicting obesity weight data. The main contribution of this research was to perform an extensive assessment of the effectiveness of LSTM-based models, including the combination of Attention-LSTM with Kalman Smoothing (KS), using two different data normalization methods (Z-score and Min-Max). This research used a publicly available dataset on obesity levels based on eating habits and physical condition, available at the UCI Machine Learning Repository. The models evaluated include the standard LSTM, Attention-LSTM, KS-LSTM, and the proposed KS-Attention-LSTM. The evaluation is conducted using the Root Mean Square Error (RMSE), the Mean Absolute Percentage Error (MAPE), and the coefficient of determination (R²). The results showed that the proposed KS-Attention-LSTM model with Min-Max normalization achieved the lowest MAPE (0.28372) and the highest R² (0.79527) among the models. This suggests that the proposed model offers advantages in terms of prediction accuracy and has a good ability to handle data variations. Therefore, the KS-Attention-LSTM model with Min-Max normalization is strongly recommended for practical implementation, particularly for time-series data prediction in the health sector. This research is beneficial and contributes an effective alternative model that improves prediction accuracy, supports decision-making in the health sector, and enriches forecasting methods. ]]>
Copyrights © 2025
