Machine Learning-Based Social Media Fake Account Detection System Using Behavioral Features

Abstract

The rapid expansion of social media platforms has significantly transformed communication, networking, and information sharing across the globe. However, this growth has also led to the proliferation of fake accounts, which are often used for malicious activities such as spreading misinformation, phishing, spamming, and identity theft. Detecting such fraudulent accounts has become a critical challenge in maintaining the integrity and security of online ecosystems. This project presents a Machine Learning-based Fake Account Detection System that analyzes user behavioral attributes to classify accounts as genuine or fake. The proposed system leverages supervised learning techniques, specifically Multinomial Naive Bayes, Linear Support Vector Classifier (SVC), and K-Nearest Neighbors (KNN), to perform classification. The system uses features such as the number of abuse reports, rejected friend requests, unaccepted friend requests, number of friends, followers, likes to unknown accounts, and comments per day. These features serve as indicators of suspicious or abnormal user behavior. The system is implemented with a user-friendly graphical interface using the Tkinter library in Python, allowing users to upload datasets, perform training and testing, and input manual data for real-time prediction. The dataset is split into training and testing sets using a randomized approach, ensuring unbiased evaluation of the models. Performance metrics such as accuracy, precision, recall, and confusion matrix are computed to evaluate each classifier's effectiveness. Experimental results demonstrate that the system can effectively distinguish between fake and genuine accounts with considerable accuracy. Visualization techniques, such as pie charts, are employed to represent model performance in terms of accuracy and error rates. Among the implemented algorithms, each has its strengths depending on the dataset characteristics, providing flexibility in choosing the best-performing model. This project contributes to enhancing cyber security by providing an efficient and scalable solution for fake account detection. The system can be extended to real-world applications in social media monitoring, fraud detection, and digital identity verification. Future improvements may include deep learning approaches and real-time data integration for improved accuracy and adaptability.