Abstract—Suicidal ideation detection on social media has become increasingly critical due to the rising prevalence of suicide-related posts. This study proposes a hybrid deep learning model to identify suicidal ideation in Indonesian-language tweets from the X platform. The model integrates Convolutional Neural Networks (CNN) for local pattern recognition, Bidirectional Gated Recurrent Units (BiGRU) for sequential context analysis, and FastText word embeddings to capture semantic nuances, especially in informal language. To further enhance performance, a Genetic Algorithm (GA) is employed for hyperparameter optimization and feature selection. The dataset comprises 50,307 annotated tweets, supplemented by 111,458 articles from the IndoNews corpus to enrich contextual understanding. Prepro- cessing steps include text cleaning, normalization, tokenization, and feature augmentation. The model was evaluated under five different experimental scenarios. Results show that the BiGRU- CNN + GA configuration in the fifth scenario achieved the highest accuracy of 86.69%, reflecting an 8.26% improvement over the baseline accuracy. These findings demonstrate the model’s effec- tiveness in detecting suicidal ideation and highlight the potential of hybrid deep learning approaches combined with evolutionary optimization in mental health-related social media analysis. The proposed approach offers a scalable solution for early detection efforts. Future work may involve real-time deployment and cross- platform validation to further enhance applicability.
Keywords—Suicide detection, CNN-BiGRU, FastText, Genetic Algorithm, X.