An Efficient Technique of Predicting Toxicity on Music Lyrics Machine Learning

Abstract

It is widely accepted that music is humanity's universal language since it can spread happiness and excitement throughout people's lives. Music is a form of art that is highly regarded worldwide. There are many ways that music lyrics affect our daily lives. In the music industry, it is crucial to prevent the reproduction of songs whose lyrics are toxic or unsuitable for children. Our mood might be impacted by listening to particularly toxic or non-toxic music. The listener's experience might be enhanced if the recommendation method eliminates toxicity. In this study, we use machine learning (ML) algorithms to classify lyrics from various musical genres and performers as toxic or non-toxic. Utilizing the Detoxify model, the toxicity score was generated and labelled the songs as toxic and non-toxic based on the scores. The study demonstrates that the configuration using the lyric data set along with TF-IDF vectorization and Ensemble of Logistic Regression, Support Vector Machine and Decision Tree as an algorithm surpasses all other designs with 94% accuracy. This classification will help the authority and policymakers of music industries to categorize the song based on the label and mention in the song description which is not appropriate for the children and set guidelines to prevent toxicity via songs.