Research Article
Supervised Learning for Suicidal Ideation Detection in Online User Content
Table 5
Comparison of different methods using different features.
| Methods | Features | Acc. | Prec. | Recall | F1-score | AUC |
| SVM | Statistics | 0.8064 | 0.8045 | 0.8189 | 0.8116 | 0.8061 | Statistics + topic | 0.8609 | 0.881 | 0.8406 | 0.8603 | 0.8613 | Statistics + topic + TF-IDF | 0.8571 | 0.8414 | 0.8865 | 0.8634 | 0.8565 | Statistics + topic + TF-IDF + POS | 0.8674 | 0.8545 | 0.8916 | 0.8727 | 0.8670 | Statistics + topic + TF-IDF + POS + LIWC | 0.9123 | 0.9144 | 0.9133 | 0.9138 | 0.9123 |
| Random Forest | Statistics | 0.7732 | 0.8094 | 0.7258 | 0.7653 | 0.7741 | Statistics + topic | 0.8973 | 0.8922 | 0.9082 | 0.9001 | 0.8971 | Statistics + topic + TF-IDF | 0.8915 | 0.8795 | 0.912 | 0.8954 | 0.8911 | Statistics + topic + TF-IDF + POS | 0.8986 | 0.8801 | 0.9273 | 0.9031 | 0.8981 | Statistics + topic + TF-IDF + POS + LIWC | 0.9357 | 0.9213 | 0.9554 | 0.938 | 0.9353 |
| GBDT | Statistics | 0.7505 | 0.7632 | 0.7398 | 0.7513 | 0.7507 | Statistics + topic | 0.898 | 0.8856 | 0.9184 | 0.9017 | 0.8976 | Statistics + topics + TF-IDF | 0.896 | 0.89 | 0.9082 | 0.899 | 0.8958 | Statistics + topic + TF-IDF + POS | 0.8928 | 0.8893 | 0.9018 | 0.8955 | 0.8926 | Statistics + topic + TF-IDF + POS + LIWC | 0.9461 | 0.9354 | 0.9605 | 0.9478 | 0.9458 |
| XGBoost | Statistics | 0.7667 | 0.7822 | 0.7513 | 0.7664 | 0.7670 | Statistics + topic | 0.8999 | 0.8938 | 0.912 | 0.9028 | 0.8997 | Statistics + topic + TF-IDF | 0.9019 | 0.8941 | 0.9158 | 0.9049 | 0.9016 | Statistics + topic + TF-IDF + POS | 0.9103 | 0.8998 | 0.9273 | 0.9133 | 0.9100 | Statistics + topic + TF-IDF + POS + LIWC | 0.9571 | 0.9499 | 0.9668 | 0.9583 | 0.9569 |
| MLFFNN | Statistics | 0.7647 | 0.7742 | 0.7742 | 0.7742 | 0.7731 | Statistics + topic | 0.8821 | 0.8740 | 0.8525 | 0.8631 | 0.8961 | Statistics + topic + TF-IDF | 0.8606 | 0.8369 | 0.8401 | 0.8385 | 0.8855 | Statistics + topic + TF-IDF + POS | 0.9068 | 0.9038 | 0.8868 | 0.8952 | 0.9369 | Statistics + topic + TF-IDF + POS + LIWC | 0.9283 | 0.9391 | 0.9205 | 0.9295 | 0.9403 |
| LSTM | word2vec word embedding | 0.9266 | 0.9786 | 0.8750 | 0.9239 | 0.9276 |
|
|