Detecting Offensive Language in English, Hindi, and Marathi using Classical Supervised Machine Learning Methods and Word/Char N-grams

In this paper, we describe our submissions for the HASOC 2021 contest. We tackled subtask 1A that addresses the problem of hate speech and offensive language identification in three languages: English, Hindi, and Marathi. We developed different models using six classical supervised machine learning methods: support vector classifier, binary support vector classifier, random forest, ada-boost classifier, multi-layer perceptron, and logistic regression. Our best submission was a model we built for offensive language identification in Marathi using random forest. This model was ranked in 6^th place out of 25 teams. Our result is lower by only 0.0059 than the result of the team that was ranked in 3^rd place. Our ML models were applied on various combinations of character and/or word n-gram features from uni-gram to 8-gram.