Comparing Hate Speech Detection Machine Learning Models Trained on Authentic and Translated Indonesian Data

Social media is growing exponentially, it has become a place where people can express themself freely. However, not often that hate speech can be found in the expressions. It is important to prevent hate speech as it is harmful for our society. There has been a lot of research about the detection of hate speech using artificial intelligence however most of this research focuses on the models, there is a few research that focuses on the data rather than the model. Our research aims to find the effects of the data itself towards the performance of the models. We achieve this by having two separate datasets: hate speech in Indonesian and hate speech in English which is translated into Indonesian using Google translate Service. We use these datasets separately into several machine learning models including Logistic Regression (LR), K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Decision Tree (DT), and Random Forest (RF). The best result was achieved by the RF model with 75% accuracy and weighted average F1 score for the Indonesian dataset and 58% accuracy and weighted average F1 score for the translated dataset. The Indonesian dataset is proven to be superior than the translated dataset.