Spam and Bot Detection Across Multiple Communication Channels Using Machine Learning

Digital communication platforms face a growing volume of spam, fraudulent content, and automated social media accounts, and the data available differ from one channel to another. This paper presents a multi-channel system that classifies Email and Short Message Service (SMS) messages as spam or legitimate (ham) and identifies suspicious automated accounts on Twitter (X). Email and SMS text is processed using Natural Language Processing (NLP) preprocessing and Term Frequency–Inverse Document Frequency (TF-IDF) features, which are classified by Random Forest models. Twitter accounts are analyzed separately from account metadata, namely follower count, following count, account age, and tweet volume, using a distinct Random Forest classifier. A FastAPI backend provides real-time prediction, batch processing, and input validation, and a web-based interface returns the predicted class, a confidence value, and dynamically generated explanations. The system reports an overall classification accuracy of 94.25% and an average prediction latency of 32 ms per request. The work shows how text-based spam classification and metadata-based bot detection can be integrated within a single application, and it identifies the evaluation details that remain to be documented.