A large proportion of online comments present on public domains are usually constructive, however a significant proportion are toxic in nature. Dataset is obtained online which are processed to remove noise from the dataset. The comments contain lot of errors which increases the number of features manifold, making the machine learning model to train the dataset by processing the dataset, in the form of transformation of raw comments before feeding it to the Classification models using a machine learning technique known as the term frequency-inverse document frequency (TF-IDF) technique. The logistic regression technique is used to train the processed dataset, which will differentiate toxic comments from non-toxic comments. The multi-headed model comprises toxicity (severetoxic, obscene, threat, insult, and identity-hate) or Non-Toxicity Evaluation, using confusion metrics for their prediction.