The proliferation of social media has increased abusive online interactions, making automated detection of abusive comments a necessity. Existing research works address similar issues in high-resource languages such as English, but there remains a gap in detecting abusive content in low-resource indic languages especially in transliterated, code mixed text using roman script. This work presents an approach to detect abusive content in transliterated Bengali. Multiple Machine Learning(ML) and Deep Learning(DL) models, such as Support Vector Machines (SVM), Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM) are applied. Our experimental results show among ML models SVM achieved highest accuracy of 85% and F1-score of 86%. CNN outperformed other DL models with an accuracy of 81% and F1-score of 80%. This study aims to enhance automated moderation systems for Bengali-speaking community. Though social media platforms have automated systems to report malicious content, such content in low- resource languages often go undetected causing extreme distress to users.