Detecting Fake Accounts Using Ensemble and Classical Machine Learning Models

Abstract

Fake accounts are a growing risk on digital platforms, as they are commonly used for spreading spam and misinformation, or to carry out fraudulent actions. Here we discuss a machine learning based technique to detect such accounts. The dataset contains 1481 real and 1337 fraudulent accounts. A detailed exploratory data analysis (EDA) was undertaken to find trends and correlations between the characteristics. Then different machine learning models were trained, including traditional models such as Logistic Regression (LR), Decision Tree (DT), K-Nearest Neighbors (KNN) and ensemble models such as Random Forest (RF), XGBoost, AdaBoost and Gradient Boosting (GB). Among the models KNN achieved the best accuracy 98.23%. GB, RF, XGBoost also achieved over 98% accuracy after tuning the models using Bayesian Optimization with Leave-One-Block-Out (LOBO), which outperforming earlier results. The study emphasizes the necessity of data preprocessing, EDA and proper model selection with tuning to enhance the reliability of the fraudulent account detection systems.