An Enhanced RNN-LSTM Model For Accurate and Real-Time Click Fraud Detection in Online

Abstract

Click fraud continues to pose a significant threat to online advertising by inflating costs and diverting budgets toward illegitimate activities, necessitating more sophisticated detection methods that go beyond traditional machine learning capabilities. In this study, we propose a robust LSTM-based Recurrent Neural Network (RNN) framework specifically designed to capture subtle behavioral patterns and time-dependent features inherent in user interaction data. By implementing a comprehensive preprocessing pipeline—comprising timestamp decomposition, feature scaling, and label encoding—we ensured high-quality input representation for our model. When benchmarked against Artificial Neural Networks (ANN) and Convolutional Neural Networks (CNN), our RNN-LSTM architecture demonstrated superior performance, achieving 99% accuracy alongside exceptional precision and recall scores. These results confirm that temporal modeling is highly effective for identifying fraudulent click sequences, positioning the LSTM model as a powerful solution for real-time ad verification and establishing a new benchmark for future advancements in intelligent fraud prevention.