Explainable Suicide Risk Prediction with DeepFusion: A Hybrid Intelligence Approach

Social media sites have emerged as a significant and trustworthy source of information, with users frequently sharing their feelings, thoughts, and opinions on them. These platforms may extract specific interests and trends from usergenerated material in real time by utilising developments in data mining technologies. In terms of gaining useful insights from such data, deep learning models have outperformed typical machine learning models. Three deep learning techniques–Deep Neural Network, Long Short Term Memory, and a CNN-LSTM hybrid model–are used in this study to estimate the likelihood that a tweet on X (previously known as twitter)may contain suicidal content.The feature extraction process relies heavily on the CNN component of the hybrid model. Using convolutional and max-pooling layers, it successfully recovers higher-level characteristics from the textual input. But the state-of-the-art natural language processing model LSTM excels at maintaining long-short-term relationships within tokens and sequences.Our research demonstrates that the CNN-LSTM hybrid model performs better than traditional deep learning models across a number of evaluation metrics, including accuracy, precision, sensitivity, and f1 score.