A Feature-engineering Approach to Machine Learning for Detecting Fake and Bot Accounts on Social Media

This paper introduces a machine learning-based classifier to identify fake accounts, bots and malevolent profiles on social networks such as X (Twitter), Facebook, and Instagram. The study combines behavioural, linguistic and network-based features to determine the genuine and automated users using the Random Forest, Support Vector Machine (SVM) and Neural Network models. As per the experimental findings, Random Forest was found to be the most balanced in terms of accuracy and interpretability with a detection accuracy of over 90 percent, whereas the Neural Networks offered more recall of complex cases. The analysis of explainability based on SHAP and LIME proved the superiority of behavioural and network predictors. The paper provides a scalable, explainable, ethically-grounded method of enhancing digital trust and platform security.