The widespread use of social networks by the general population has led to growing instances of hate speech propagation on these networks. The toxic language found in these instances has been known to adversely affect the mental well-being of the people that these words are aimed at. Toxicity analysis is a distinct kind of sentiment analysis task aimed at determining the toxicity level or level of hatred found in a piece of text. It utilises lexical or deep learning methods to classify the text into its correct toxic category. In light of this, this paper proposes a novel method that models social media texts (such as tweets) as multi-layer graphs and employs Graph Neural Networks to classify them with the correct toxicity label. Further, to counter the issue of class imbalance in existing datasets, we create augmented versions of pre-existing datasets using the GPT-2 model and provide them for open access.