Cyberbullying and Suicide Ideation Detection via Hybrid Machine Learning Model

Increase in Cyberbullying and suicide ideas has led a great need in its ideation detection to prevent it. Since the detection of it is even difficult for human and hence it is much more complicated in terms of automatic detection. Many past efforts in this field were tedious, few lexicon-based methods were functional but later turned out to be scope-limiting. The current requirement emphasizes on the use of Machine learning models for different Natural Language Processing problems. Use of these technology gives the quicker and accurate way in detection. In this paper SVM, Random-forest, BDRNN GRU AND BDRNN LSTM are used as convolutional networks. The model uses pre-trained glove embedding layer and hybrid BiGRU model that gives the accuracy of 92.70% and F-score of 0.9431 for cyberbullying detection and accuracy of 97% for suicide ideation. Based on these results the performance of this model is tested by the analysis of four parameters which are Recall, Precision, $\mathrm{F}_{\mathrm{m}\mathrm{e}\mathrm{a}\mathrm{s}\mathrm{u}\mathrm{r}\mathrm{e}}$ and Accuracy.