Fake Instagram Account Detection Using Single and Stacking Ensemble Models on Integrated Static and Temporal Features

Abstract

This research examines the detection of fake Instagram accounts by comparing single machine learning classifiers with a stacking ensemble model applied to integrated static and temporal features. A dataset of 7,594 accounts was collected over five months using multiple scraping tools, ensuring a balanced representation of real and fake profiles. Key behavioral, engagement, and account-level features were extracted to capture users’ activity patterns. Among the individual classifiers, Random Forest achieved the highest performance; however, the stacking ensemble model outperformed all single models, reaching an accuracy of 0.9888 with superior precision, recall, and F-measure values. The findings demonstrate that integrating temporal metrics with ensemble learning significantly enhances detection accuracy and provides a more reliable approach for identifying fraudulent accounts on social media platforms.