In the past few years, there has been an expeditious growth in the usage of social media platforms and blogging websites which has passed 3.8 billion marks of active users that use text as a prominent
means for interactive communication. A fraction of users spread misinformation on social media. As Twitter has 330 million monthly active users, researchers have been using it as a source of data for misinformation identification. In this paper, we have proposed a Twitter dataset for fine-grained classification. Our dataset is consists of 1970 manually annotated tweets and is categorized into 4 misinformation classes, i.e, “Irrelevant”, “Conspiracy”, “True Information”, and “False Information” based on response erupted during COVID-19. In this work, we also generated useful insights on our dataset and performed a systematic analysis of various language models, namely, RNN (BiLSTM, LSTM), CNN (TextCNN), BERT, ROBERTA, and ALBERT for the classification task on our dataset. Through our work, we aim at contributing to the substantial efforts of the research community for the identification and mitigation of misinformation on the internet.