Performance Comparison of Machine Learning and Deep Learning Algorithms in Detecting Online Hate Speech

The main objective of this research is to analyze and compare the performance of machine learning (ML) and deep learning (DL) algorithms in detecting online hate speech. Therefore, Support Vector Machine (SVM), Random Forest (RF), Decision Tree (DT), Logistic Regression (LR), Convolution Neural Network (CNN), Recurrent Neural Network_Long Short-Term Memory (RNN_LSTM), BERT (Bidirectional Encoder Representations from Transformers), and Distil BERT algorithms have been explored and analyzed in this research. This research has applied the dataset on hate speech which was developed by Andry Samoshyn which is publicly available in Kaggle. ML algorithms and DL algorithms have got good scores in accuracy. In ML, SVM, RF, and LR have got top accuracy values. In DL algorithms, RNN_LSTM, Distil BERT, and BERT have performed well in accuracy. Based on F-measurement, DL classifiers have outperformed ML algorithms. Distil BERT has obtained the highest F-measurement scores. When we compare the overall performances, DL is performed well rather than ML in detecting hate speech. Especially transformer-based models of DL are more efficient than other DL and ML algorithms.