<![CDATA[Early stress detection through digital conversational text is crucial for mental health, but research in Indonesian is still limited. This study designs and evaluates a Long Short-Term Memory (LSTM)-based deep learning model to classify Indonesian text as stressful or non-stressful. The model was trained and tested using a labeled dataset of 11,000 samples. The methodology included text preprocessing, model training, and sensitivity analysis of hyperparameters such as learning rate, batch size, and number of LSTM units to find the optimal configuration. The proposed model demonstrated strong performance with an accuracy of 86.48% and a balanced F1-Score of 0.87 (non-stress) and 0.86 (stress), outperforming several previous baselines. Training curve analysis identified clear overfitting, while hyperparameter sensitivity analysis revealed that the default configuration with 64 LSTM units was suboptimal—performance improved with the use of 128 LSTM units or a batch size of 128. This study confirms the effectiveness of LSTM for stress detection in Indonesian text, while also demonstrating the need for further hyperparameter optimization and the need for more robust overfitting handling techniques.]]>
Copyrights © 2026
