One of the most challenging tasks while processing natural language text is to authenticate the correctness of the provided information particularly for classification of fake news. Fake news is a growing source of apprehension in recent times for hate speech as well. For instance, the followers of various beliefs face constant discrimination and receive negative perspectives directed at them. Fake news is one of the most prominent reasons for various kinds of racism and stands at par with individual, interpersonal, and structural racism types observed worldwide yet it does not get much importance and remains to be neglected. In this paper, to mitigate racism, we address the fake news regarding beliefs related to Islam as a case study. Though fake news remained to be a concerning factor since the beginning of Islam, a significant increase has been noticed in it for the last three years. Additionally, the accessibility of social media platforms and the growth in their use have helped to propagate misinformation, hate speech, and unfavorable views about Islam. Based on these deductions, this study intends to categorize such anti-Islamic content and misinformation found in Twitter posts. Several preprocessing and data enhancement steps were employed on retrieved data. Word2vec and GloVe were implemented to derive deep features while TF-IDF and BOW were applied to derive textual features from the data respectively. Finally, the classification phase was performed using four Machine-based predictive analysis (ML) algorithms Random Forest (RF), Naïve Bayes (NB), Logistic Regression (LR), Support Vector Machine (SVM), and a custom deep CNN. The results when compared with certain performance evaluation measures show that on average, ML-models perform better than the CNN for the utilized dataset.